plateform
stringclasses 1
value | repo_name
stringlengths 13
113
| name
stringlengths 3
74
| ext
stringclasses 1
value | path
stringlengths 12
229
| size
int64 23
843k
| source_encoding
stringclasses 9
values | md5
stringlengths 32
32
| text
stringlengths 23
843k
|
|---|---|---|---|---|---|---|---|---|
github
|
happyharrycn/unsupervised_edges-master
|
edgesEvalImgFast.m
|
.m
|
unsupervised_edges-master/structured_edges/edgesEvalImgFast.m
| 4,994 |
utf_8
|
3470ed4119e98435cb12f7c47902934e
|
function [thrs,cntR,sumR,cntP,sumP,V] = edgesEvalImgFast( E, G, varargin )
% Calculate edge precision/recall results for single edge image.
%
% Enhanced replacement for evaluation_bdry_image() from BSDS500 code:
% http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/
% Uses same format and is fully compatible with evaluation_bdry_image.
% Given default prms results are *identical* to evaluation_bdry_image.
%
% In addition to performing the evaluation, this function can optionally
% create a visualization of the matches and errors for a given edge result.
% The visualization of edge matches V has the following color coding:
% green=true positive, blue=false positive, red=false negative
% If multiple ground truth labels are given the false negatives have
% varying strength (and true positives can match *any* ground truth).
%
% This function calls the mex file correspondPixels. Pre-compiled binaries
% for some systems are provided in /private, source for correspondPixels is
% available as part of the BSDS500 dataset (see link above). Note:
% correspondPixels is computationally expensive and very slow in practice.
%
% USAGE
% [thrs,cntR,sumR,cntP,sumP,V] = edgesEvalImg( E, G, [prms] )
%
% INPUTS
% E - [h x w] edge probability map (may be a filename)
% G - file containing a cell of ground truth boundaries
% prms - parameters (struct or name/value pairs)
% .out - [''] optional output file for writing results
% .thrs - [99] number or vector of thresholds for evaluation
% .maxDist - [.0075] maximum tolerance for edge match
% .thin - [1] if true thin boundary maps
% .scale = [1] resize the gt and image for evalution
%
% OUTPUTS
% thrs - [Kx1] vector of threshold values
% cntR,sumR - [Kx1] ratios give recall per threshold
% cntP,sumP - [Kx1] ratios give precision per threshold
% V - [hxwx3xK] visualization of edge matches
%
% EXAMPLE
%
% See also edgesEvalDirFast
%
% Structured Edge Detection Toolbox Version 3.01
% Code written by Piotr Dollar, 2014. Modified by Yin Li
% Licensed under the MSR-LA Full Rights License [see license.txt]
% get additional parameters
dfs={ 'out','', 'thrs',39, 'maxDist',.0075, 'thin',1 , 'scale', 1.0};
[out,thrs,maxDist,thin, scale] = getPrmDflt(varargin,dfs,1);
if(any(mod(thrs,1)>0)), K=length(thrs); thrs=thrs(:); else
K=thrs; thrs=linspace(1/(K+1),1-1/(K+1),K)'; end
% load edges (E) and ground truth (G)
if(all(ischar(E))),
E=double(imread(E))/255;
if (scale-1.0) > eps
E=imresize(E, scale, 'nearest');
end
end
G=load(G); G=G.groundTruth; n=length(G);
for g=1:n,
G{g}=double(G{g}.Boundaries);
if abs(scale-1.0) > eps
G{g}=imresize(G{g}, scale, 'nearest');
end
end
% evaluate edge result at each threshold
Z=zeros(K,1); cntR=Z; sumR=Z; cntP=Z; sumP=Z;
if(nargout>=6), V=zeros([size(E) 3 K]); end
for k = 1:K
% threshhold and thin E
E1 = double(E>=max(eps,thrs(k)));
if(thin), E1=double(bwmorph(E1,'thin',inf)); end
% compare to each ground truth in turn and accumualte
Z=zeros(size(E)); matchE=Z; matchG=Z; allG=Z;
for g = 1:n
[matchE1,matchG1] = correspondPixels(E1,G{g},maxDist);
matchE = matchE | matchE1>0;
matchG = matchG + double(matchG1>0);
allG = allG + G{g};
end
% compute recall (summed over each gt image)
cntR(k) = sum(matchG(:)); sumR(k) = sum(allG(:));
% compute precision (edges can match any gt image)
cntP(k) = nnz(matchE); sumP(k) = nnz(E1);
% optinally create visualization of matches
if(nargout<6), continue; end; cs=[1 0 0; 0 .7 0; .7 .8 1]; cs=cs-1;
FP=E1-matchE; TP=matchE; FN=(allG-matchG)/n;
for g=1:3, V(:,:,g,k)=max(0,1+FN*cs(1,g)+TP*cs(2,g)+FP*cs(3,g)); end
V(:,2:end,:,k) = min(V(:,2:end,:,k),V(:,1:end-1,:,k));
V(2:end,:,:,k) = min(V(2:end,:,:,k),V(1:end-1,:,:,k));
end
% if output file specified write results to disk
[thrs2, cntR2, sumR2, cntP2, sumP2] = quadInterpolate(thrs, cntR, sumR, cntP, sumP, 100);
if(isempty(out)), return; end; fid=fopen(out,'w'); assert(fid~=1);
fprintf(fid,'%10g %10g %10g %10g %10g\n',[thrs2 cntR2 sumR2 cntP2 sumP2]');
fclose(fid);
end
% interpolate the PR curve by quadratic function
function [thrs2, cntR2, sumR2, cntP2, sumP2] = quadInterpolate(thrs, cntR, sumR, cntP, sumP,K)
thrs2 = linspace(0.01,1,K)';thrs2 = [0.005;thrs2];
sumR2 = interp1(thrs, sumR, thrs2, 'spline'); sumR2 = max(round(sumR2),0);
sumP2 = interp1(thrs, sumP, thrs2, 'spline');
[~, ind] = min(sumP2); sumP2(ind+1:end)=0;sumP2 = max(round(sumP2),0);
R = cntR./(sumR+eps); R2 = interp1(thrs, R, thrs2,'spline');[~, ind] = min(R2);R2(ind+1:end)=0;
R2 = min(R2,1);R2 = max(R2,0);
cntR2 = R2.*sumR2;cntR2 = max(round(cntR2),0);
P = cntP./(sumP+eps);
try
P2 = interp1(thrs, P, thrs2, 'spline');
%P2 = spline(thrs, P, thrs2);
catch
print('oops\n');
P2 = interp1(thrs, P, thrs2, 'linear');
end
P2 = min(P2,1);P2 = max(P2,0);
cntP2 = P2.*sumP2; cntP2 = max(round(cntP2),0);
end
|
github
|
happyharrycn/unsupervised_edges-master
|
edgesEvalDir.m
|
.m
|
unsupervised_edges-master/structured_edges/edgesEvalDir.m
| 5,852 |
utf_8
|
b708b92045eaa75fa68d09e169447bb6
|
function varargout = edgesEvalDir( varargin )
% Calculate edge precision/recall results for directory of edge images.
%
% Enhanced replacement for boundaryBench() from BSDS500 code:
% http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/
% Uses same format for results and is fully compatible with boundaryBench.
% Given default prms results are *identical* to boundaryBench with the
% additional 9th output of R50 (recall at 50% precision).
%
% The odsF/P/R/T are results at the ODS (optimal dataset scale).
% The oisF/P/R are results at the OIS (optimal image scale).
% Naming convention: P=precision, R=recall, F=2/(1/P+1/R), T=threshold.
%
% In addition to the outputs, edgesEvalDir() creates three files:
% eval_bdry_img.txt - per image OIS results [imgId T R P F]
% eval_bdry_thr.txt - per threshold ODS results [T R P F]
% eval_bdry.txt - complete results (*re-ordered* copy of output)
% These files are identical to the ones created by boundaryBench.
%
% USAGE
% [odsF,odsP,odsR,odsT,oisF,oisP,oisR,AP,R50] = edgesEvalDir( prms )
% [ODS,~,~,~,OIS,~,~,AP,R50] = edgesEvalDir( prms )
%
% INPUTS
% prms - parameters (struct or name/value pairs)
% .resDir - ['REQ'] dir containing edge detection results (.png)
% .gtDir - ['REQ'] dir containing ground truth (.mat)
% .pDistr - [{'type','parfor'}] parameters for fevalDistr
% .cleanup - [0] if true delete temporary files
% .thrs - [99] number or vector of thresholds for evaluation
% .maxDist - [.0075] maximum tolerance for edge match
% .thin - [1] if true thin boundary maps
%
% OUTPUTS
% odsF/P/R/T - F-measure, precision, recall and threshold at ODS
% oisF/P/R - F-measure, precision, and recall at OIS
% AP - average precision
% R50 - recall at 50% precision
%
% EXAMPLE
%
% See also edgesEvalImg, edgesEvalPlot
%
% Structured Edge Detection Toolbox Version 3.01
% Code written by Piotr Dollar, 2014.
% Licensed under the MSR-LA Full Rights License [see license.txt]
% get additional parameters
dfs={ 'resDir','REQ', 'gtDir','REQ', 'pDistr',{{'type','parfor'}}, ...
'cleanup',0, 'thrs',99, 'maxDist',.0075, 'thin',1 };
p=getPrmDflt(varargin,dfs,1); resDir=p.resDir; gtDir=p.gtDir;
evalDir=[resDir '-eval/']; if(~exist(evalDir,'dir')), mkdir(evalDir); end
% check if results already exist, if so load and return
fNm = fullfile(evalDir,'eval_bdry.txt');
if(exist(fNm,'file')), R=dlmread(fNm); R=mat2cell2(R,[1 8]);
varargout=R([4 3 2 1 7 6 5 8]); if(nargout<=8), return; end;
R=dlmread(fullfile(evalDir,'eval_bdry_thr.txt')); P=R(:,3); R=R(:,2);
[~,o]=unique(P); R50=interp1(P(o),R(o),max(P(o(1)),.5));
varargout=[varargout R50]; return;
end
% perform evaluation on each image (this part can be very slow)
assert(exist(resDir,'dir')==7); assert(exist(gtDir,'dir')==7);
ids=dir(fullfile(gtDir,'*.mat')); ids={ids.name}; n=length(ids);
do=false(1,n); jobs=cell(1,n); res=cell(1,n);
for i=1:n, id=ids{i}(1:end-4);
res{i}=fullfile(evalDir,[id '_ev1.txt']); do(i)=~exist(res{i},'file');
im1=fullfile(resDir,[id '.png']); gt1=fullfile(gtDir,[id '.mat']);
jobs{i}={im1,gt1,'out',res{i},'thrs',p.thrs,'maxDist',p.maxDist,...
'thin',p.thin}; if(0), edgesEvalImg(jobs{i}{:}); end
end
fevalDistr('edgesEvalImg',jobs(do),p.pDistr{:});
% collect evaluation results
I=dlmread(res{1}); T=I(:,1);
Z=zeros(numel(T),1); cntR=Z; sumR=Z; cntP=Z; sumP=Z;
oisCntR=0; oisSumR=0; oisCntP=0; oisSumP=0; scores=zeros(n,5);
for i=1:n
% load image results and accumulate
I = dlmread(res{i});
cntR1=I(:,2); cntR=cntR+cntR1; sumR1=I(:,3); sumR=sumR+sumR1;
cntP1=I(:,4); cntP=cntP+cntP1; sumP1=I(:,5); sumP=sumP+sumP1;
% compute OIS scores for image
[R,P,F] = computeRPF(cntR1,sumR1,cntP1,sumP1); [~,k]=max(F);
[oisR1,oisP1,oisF1,oisT1] = findBestRPF(T,R,P);
scores(i,:) = [i oisT1 oisR1 oisP1 oisF1];
% oisCnt/Sum will be used to compute dataset OIS scores
oisCntR=oisCntR+cntR1(k); oisSumR=oisSumR+sumR1(k);
oisCntP=oisCntP+cntP1(k); oisSumP=oisSumP+sumP1(k);
end
% compute ODS R/P/F and OIS R/P/F
[R,P,F] = computeRPF(cntR,sumR,cntP,sumP);
[odsR,odsP,odsF,odsT] = findBestRPF(T,R,P);
[oisR,oisP,oisF] = computeRPF(oisCntR,oisSumR,oisCntP,oisSumP);
% compute AP/R50 (interpolating 100 values, has minor bug: should be /101)
if(0), R=[0; R; 1]; P=[1; P; 0]; F=[0; F; 0]; T=[1; T; 0]; end
[~,k]=unique(R); k=k(end:-1:1); R=R(k); P=P(k); T=T(k); F=F(k); AP=0;
if(numel(R)>1), AP=interp1(R,P,0:.01:1); AP=sum(AP(~isnan(AP)))/100; end
[~,o]=unique(P); R50=interp1(P(o),R(o),max(P(o(1)),.5));
% write results to disk
varargout = {odsF,odsP,odsR,odsT,oisF,oisP,oisR,AP,R50};
writeRes(evalDir,'eval_bdry_img.txt',scores);
writeRes(evalDir,'eval_bdry_thr.txt',[T R P F]);
writeRes(evalDir,'eval_bdry.txt',[varargout{[4 3 2 1 7 6 5 8]}]);
% optionally perform cleanup
if( p.cleanup ), delete([evalDir '/*_ev1.txt']);
delete([resDir '/*.png']); rmdir(resDir); end
end
function [R,P,F] = computeRPF(cntR,sumR,cntP,sumP)
% compute precision, recall and F measure given cnts and sums
R=cntR./max(eps,sumR); P=cntP./max(eps,sumP); F=2*P.*R./max(eps,P+R);
end
function [bstR,bstP,bstF,bstT] = findBestRPF(T,R,P)
% linearly interpolate to find best thr for optimizing F
if(numel(T)==1), bstT=T; bstR=R; bstP=P;
bstF=2*P.*R./max(eps,P+R); return; end
A=linspace(0,1,100); B=1-A; bstF=-1;
for j = 2:numel(T)
Rj=R(j).*A+R(j-1).*B; Pj=P(j).*A+P(j-1).*B; Tj=T(j).*A+T(j-1).*B;
Fj=2.*Pj.*Rj./max(eps,Pj+Rj); [f,k]=max(Fj);
if(f>bstF), bstT=Tj(k); bstR=Rj(k); bstP=Pj(k); bstF=f; end
end
end
function writeRes( alg, fNm, vals )
% write results to disk
k=size(vals,2); fNm=fullfile(alg,fNm); fid=fopen(fNm,'w');
if(fid==-1), error('Could not open file %s for writing.',fNm); end
frmt=repmat('%10g ',[1 k]); frmt=[frmt(1:end-1) '\n'];
fprintf(fid,frmt,vals'); fclose(fid);
end
|
github
|
happyharrycn/unsupervised_edges-master
|
edgesEvalDirFast.m
|
.m
|
unsupervised_edges-master/structured_edges/edgesEvalDirFast.m
| 5,998 |
utf_8
|
2705468968d13cfb5879826f8fc1a673
|
function varargout = edgesEvalDirFast( varargin )
% Calculate edge precision/recall results for directory of edge images.
%
% Enhanced replacement for boundaryBench() from BSDS500 code:
% http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/
% Uses same format for results and is fully compatible with boundaryBench.
% Given default prms results are *identical* to boundaryBench with the
% additional 9th output of R50 (recall at 50% precision).
%
% The odsF/P/R/T are results at the ODS (optimal dataset scale).
% The oisF/P/R are results at the OIS (optimal image scale).
% Naming convention: P=precision, R=recall, F=2/(1/P+1/R), T=threshold.
%
% In addition to the outputs, edgesEvalDir() creates three files:
% eval_bdry_img.txt - per image OIS results [imgId T R P F]
% eval_bdry_thr.txt - per threshold ODS results [T R P F]
% eval_bdry.txt - complete results (*re-ordered* copy of output)
% These files are identical to the ones created by boundaryBench.
%
% USAGE
% [odsF,odsP,odsR,odsT,oisF,oisP,oisR,AP,R50] = edgesEvalDir( prms )
% [ODS,~,~,~,OIS,~,~,AP,R50] = edgesEvalDir( prms )
%
% INPUTS
% prms - parameters (struct or name/value pairs)
% .resDir - ['REQ'] dir containing edge detection results (.png)
% .gtDir - ['REQ'] dir containing ground truth (.mat)
% .pDistr - [{'type','parfor'}] parameters for fevalDistr
% .cleanup - [0] if true delete temporary files
% .thrs - [99] number or vector of thresholds for evaluation
% .maxDist - [.0075] maximum tolerance for edge match
% .thin - [1] if true thin boundary maps
%
% OUTPUTS
% odsF/P/R/T - F-measure, precision, recall and threshold at ODS
% oisF/P/R - F-measure, precision, and recall at OIS
% AP - average precision
% R50 - recall at 50% precision
%
% EXAMPLE
%
% See also edgesEvalImg, edgesEvalPlot
%
% Structured Edge Detection Toolbox Version 3.01
% Code written by Piotr Dollar, 2014. Modified by Yin Li
% Licensed under the MSR-LA Full Rights License [see license.txt]
% get additional parameters
dfs={ 'resDir','REQ', 'gtDir','REQ', 'pDistr',{{'type','parfor'}}, ...
'cleanup',0, 'thrs',39, 'maxDist',.0075, 'thin',1, 'scale', 1 };
p=getPrmDflt(varargin,dfs,1); resDir=p.resDir; gtDir=p.gtDir;
evalDir=[resDir '-eval/']; if(~exist(evalDir,'dir')), mkdir(evalDir); end
% check if results already exist, if so load and return
fNm = fullfile(evalDir,'eval_bdry.txt');
if(exist(fNm,'file')), R=dlmread(fNm); R=mat2cell2(R,[1 8]);
varargout=R([4 3 2 1 7 6 5 8]); if(nargout<=8), return; end;
R=dlmread(fullfile(evalDir,'eval_bdry_thr.txt')); P=R(:,3); R=R(:,2);
[~,o]=unique(P); R50=interp1(P(o),R(o),max(P(o(1)),.5));
varargout=[varargout R50]; return;
end
% perform evaluation on each image (this part can be very slow)
assert(exist(resDir,'dir')==7); assert(exist(gtDir,'dir')==7);
ids=dir(fullfile(gtDir,'*.mat')); ids={ids.name}; n=length(ids);
do=false(1,n); jobs=cell(1,n); res=cell(1,n);
for i=1:n, id=ids{i}(1:end-4);
res{i}=fullfile(evalDir,[id '_ev1.txt']); do(i)=~exist(res{i},'file');
im1=fullfile(resDir,[id '.png']); gt1=fullfile(gtDir,[id '.mat']);
if exist(im1, 'file') && exist(gt1, 'file')
jobs{i}={im1,gt1,'out',res{i},'thrs',p.thrs,'maxDist',p.maxDist,...
'thin',p.thin, 'scale', p.scale};
else
do(i) = false;
fprintf('Warning: %s missing!\n', im1);
end
end
fevalDistr('edgesEvalImgFast',jobs(do),p.pDistr{:});
% collect evaluation results
I=dlmread(res{1}); T=I(:,1);
Z=zeros(numel(T),1); cntR=Z; sumR=Z; cntP=Z; sumP=Z;
oisCntR=0; oisSumR=0; oisCntP=0; oisSumP=0; scores=zeros(n,5);
for i=1:n
% load image results and accumulate
I = dlmread(res{i});
cntR1=I(:,2); cntR=cntR+cntR1; sumR1=I(:,3); sumR=sumR+sumR1;
cntP1=I(:,4); cntP=cntP+cntP1; sumP1=I(:,5); sumP=sumP+sumP1;
% compute OIS scores for image
[R,P,F] = computeRPF(cntR1,sumR1,cntP1,sumP1); [~,k]=max(F);
[oisR1,oisP1,oisF1,oisT1] = findBestRPF(T,R,P);
scores(i,:) = [i oisT1 oisR1 oisP1 oisF1];
% oisCnt/Sum will be used to compute dataset OIS scores
oisCntR=oisCntR+cntR1(k); oisSumR=oisSumR+sumR1(k);
oisCntP=oisCntP+cntP1(k); oisSumP=oisSumP+sumP1(k);
end
% compute ODS R/P/F and OIS R/P/F
[R,P,F] = computeRPF(cntR,sumR,cntP,sumP);
[odsR,odsP,odsF,odsT] = findBestRPF(T,R,P);
[oisR,oisP,oisF] = computeRPF(oisCntR,oisSumR,oisCntP,oisSumP);
% compute AP/R50 (interpolating 100 values, has minor bug: should be /101)
if(0), R=[0; R; 1]; P=[1; P; 0]; F=[0; F; 0]; T=[1; T; 0]; end
[~,k]=unique(R); k=k(end:-1:1); R=R(k); P=P(k); T=T(k); F=F(k); AP=0;
if(numel(R)>1), AP=interp1(R,P,0:.01:1); AP=sum(AP(~isnan(AP)))/100; end
[~,o]=unique(P); R50=interp1(P(o),R(o),max(P(o(1)),.5));
% write results to disk
varargout = {odsF,odsP,odsR,odsT,oisF,oisP,oisR,AP,R50};
writeRes(evalDir,'eval_bdry_img.txt',scores);
writeRes(evalDir,'eval_bdry_thr.txt',[T R P F]);
writeRes(evalDir,'eval_bdry.txt',[varargout{[4 3 2 1 7 6 5 8]}]);
% optionally perform cleanup
if( p.cleanup ), delete([evalDir '/*_ev1.txt']);
delete([resDir '/*.png']); rmdir(resDir); end
end
function [R,P,F] = computeRPF(cntR,sumR,cntP,sumP)
% compute precision, recall and F measure given cnts and sums
R=cntR./max(eps,sumR); P=cntP./max(eps,sumP); F=2*P.*R./max(eps,P+R);
end
function [bstR,bstP,bstF,bstT] = findBestRPF(T,R,P)
% linearly interpolate to find best thr for optimizing F
if(numel(T)==1), bstT=T; bstR=R; bstP=P;
bstF=2*P.*R./max(eps,P+R); return; end
A=linspace(0,1,100); B=1-A; bstF=-1;
for j = 2:numel(T)
Rj=R(j).*A+R(j-1).*B; Pj=P(j).*A+P(j-1).*B; Tj=T(j).*A+T(j-1).*B;
Fj=2.*Pj.*Rj./max(eps,Pj+Rj); [f,k]=max(Fj);
if(f>bstF), bstT=Tj(k); bstR=Rj(k); bstP=Pj(k); bstF=f; end
end
end
function writeRes( alg, fNm, vals )
% write results to disk
k=size(vals,2); fNm=fullfile(alg,fNm); fid=fopen(fNm,'w');
if(fid==-1), error('Could not open file %s for writing.',fNm); end
frmt=repmat('%10g ',[1 k]); frmt=[frmt(1:end-1) '\n'];
fprintf(fid,frmt,vals'); fclose(fid);
end
|
github
|
happyharrycn/unsupervised_edges-master
|
edgesTrain.m
|
.m
|
unsupervised_edges-master/structured_edges/edgesTrain.m
| 16,041 |
utf_8
|
3df2375071950b7d0bbce36c6111524e
|
function model = edgesTrain( trnImgDir, trnGtDir, varargin )
% Train structured edge detector.
%
% For an introductory tutorial please see edgesDemo.m.
%
% USAGE
% opts = edgesTrain()
% model = edgesTrain( trnImgDir, trnGtDir, opts )
%
% INPUTS
% trnImgDir - folder with all training images
% trnGtDir - folder with all edge maps
% opts - parameters (struct or name/value pairs)
% (1) model parameters:
% .imWidth - [32] width of image patches
% .gtWidth - [16] width of ground truth patches
% (2) tree parameters:
% .nPos - [5e5] number of positive patches per tree
% .nNeg - [5e5] number of negative patches per tree
% .nImgs - [inf] maximum number of images to use for training
% .nTrees - [8] number of trees in forest to train
% .fracFtrs - [1/4] fraction of features to use to train each tree
% .minCount - [1] minimum number of data points to allow split
% .minChild - [8] minimum number of data points allowed at child nodes
% .maxDepth - [64] maximum depth of tree
% .discretize - ['pca'] options include 'pca' and 'kmeans'
% .nSamples - [256] number of samples for clustering structured labels
% .nClasses - [2] number of classes (clusters) for binary splits
% .split - ['gini'] options include 'gini', 'entropy' and 'twoing'
% (3) feature parameters:
% .nOrients - [4] number of orientations per gradient scale
% .grdSmooth - [0] radius for image gradient smoothing (using convTri)
% .chnSmooth - [2] radius for reg channel smoothing (using convTri)
% .simSmooth - [8] radius for sim channel smoothing (using convTri)
% .normRad - [4] gradient normalization radius (see gradientMag)
% .shrink - [2] amount to shrink channels
% .nCells - [5] number of self similarity cells
% .rgbd - [0] 0:RGB, 1:depth, 2:RBG+depth (for NYU data only)
% (4) detection parameters (can be altered after training):
% .stride - [2] stride at which to compute edges
% .multiscale - [0] if true run multiscale edge detector
% .sharpen - [2] sharpening amount (can only decrease after training)
% .nTreesEval - [4] number of trees to evaluate per location
% .nThreads - [4] number of threads for evaluation of trees
% .nms - [0] if true apply non-maximum suppression to edges
% (5) other parameters:
% .seed - [1] seed for random stream (for reproducibility)
% .useParfor - [0] if true train trees in parallel (memory intensive)
% .modelDir - ['models/'] target directory for storing models
% .modelFnm - ['model'] model filename
% .bsdsDir - ['BSR/BSDS500/data/'] location of BSDS dataset
% .scale = [1] scale factor for gt and images
%
% OUTPUTS
% model - trained structured edge detector w the following fields
% .opts - input parameters and constants
% .thrs - [nNodes x nTrees] threshold corresponding to each fid
% .fids - [nNodes x nTrees] feature ids for each node
% .child - [nNodes x nTrees] index of child for each node
% .count - [nNodes x nTrees] number of data points at each node
% .depth - [nNodes x nTrees] depth of each node
% .eBins - data structure for storing all node edge maps
% .eBnds - data structure for storing all node edge maps
%
% EXAMPLE
%
% See also edgesDemo, edgesChns, edgesDetect, forestTrain
%
% Structured Edge Detection Toolbox Version 3.01
% Code written by Piotr Dollar, 2014. Modified by Yin
% Licensed under the MSR-LA Full Rights License [see license.txt]
% get default parameters
% I have hard coded some param here
dfs={'imWidth',32, 'gtWidth',16, 'nPos', 5e5, 'nNeg', 5e5, 'nImgs',inf, ...
'nTrees',8, 'fracFtrs',1/4, 'minCount',1, 'minChild',8, ...
'maxDepth',64, 'discretize','pca', 'nSamples',256, 'nClasses',2, ...
'split','gini', 'nOrients',4, 'grdSmooth',0, 'chnSmooth',2, ...
'simSmooth',8, 'normRad',4, 'shrink',2, 'nCells',5, 'rgbd',0, ...
'stride',2, 'multiscale',1, 'sharpen',2, 'nTreesEval',4, ...
'nThreads', 4, 'nms',0, 'seed',1, 'useParfor', 1, 'modelDir','./tmp/', ...
'modelFnm','model', 'scale', 1, 'nGt', 4, 'threshBracket', [0.1 0.4]};
opts = getPrmDflt(varargin,dfs,1);
if(nargin==0), model=opts; return; end
% if forest exists load it and return
% modelDir -> param.tmpFolder
% modelFnm -> param.dataset + param.iter
cd(fileparts(mfilename('fullpath')));
forestDir = [opts.modelDir '/forest/'];
forestFn = [forestDir opts.modelFnm];
if(exist([forestFn '.mat'], 'file'))
load([forestFn '.mat']); return; end
% compute constants and store in opts
nTrees=opts.nTrees; nCells=opts.nCells; shrink=opts.shrink;
opts.nPos=round(opts.nPos); opts.nNeg=round(opts.nNeg);
opts.nTreesEval=min(opts.nTreesEval,nTrees);
opts.stride=max(opts.stride,shrink);
imWidth=opts.imWidth; gtWidth=opts.gtWidth;
imWidth=round(max(gtWidth,imWidth)/shrink/2)*shrink*2;
opts.imWidth=imWidth; opts.gtWidth=gtWidth;
nChnsGrad=(opts.nOrients+1)*2; nChnsColor=3;
nChns = nChnsGrad+nChnsColor; opts.nChns = nChns;
opts.nChnFtrs = imWidth*imWidth*nChns/shrink/shrink;
opts.nSimFtrs = (nCells*nCells)*(nCells*nCells-1)/2*nChns;
opts.nTotFtrs = opts.nChnFtrs + opts.nSimFtrs; disp(opts);
% generate stream for reproducibility of model
stream=RandStream('mrg32k3a','Seed',opts.seed);
% train nTrees random trees (can be trained with parfor if enough memory)
if(opts.useParfor),
parfor i=1:nTrees,
trainTree(trnImgDir, trnGtDir, opts,stream,i);
end
else
for i=1:nTrees,
trainTree(trnImgDir, trnGtDir, opts,stream,i);
end
end
% merge trees and save model
model = mergeTrees( opts );
if(~exist(forestDir,'dir')), mkdir(forestDir); end
save([forestFn '.mat'], 'model', '-v7.3');
end
%% merging all trees into forest
function model = mergeTrees( opts )
% accumulate trees and merge into final model
nTrees=opts.nTrees; gtWidth=opts.gtWidth;
treeFn = [opts.modelDir '/tree/' opts.modelFnm '_tree'];
for i=1:nTrees
t=load([treeFn int2str2(i,3) '.mat'],'tree'); t=t.tree;
if(i==1), trees=t(ones(1,nTrees)); else trees(i)=t; end
end
nNodes=0; for i=1:nTrees, nNodes=max(nNodes,size(trees(i).fids,1)); end
% merge all fields of all trees
model.opts=opts; Z=zeros(nNodes,nTrees,'uint32');
model.thrs=zeros(nNodes,nTrees,'single');
model.fids=Z; model.child=Z; model.count=Z; model.depth=Z;
model.segs=zeros(gtWidth,gtWidth,nNodes,nTrees,'uint8');
for i=1:nTrees, tree=trees(i); nNodes1=size(tree.fids,1);
model.fids(1:nNodes1,i) = tree.fids;
model.thrs(1:nNodes1,i) = tree.thrs;
model.child(1:nNodes1,i) = tree.child;
model.count(1:nNodes1,i) = tree.count;
model.depth(1:nNodes1,i) = tree.depth;
model.segs(:,:,1:nNodes1,i) = tree.hs-1;
end
% remove very small segments (<=5 pixels)
segs=model.segs; nSegs=squeeze(max(max(segs)))+1;
parfor i=1:nTrees*nNodes, m=nSegs(i);
if(m==1), continue; end; S=segs(:,:,i); del=0;
for j=1:m, Sj=(S==j-1); if(nnz(Sj)>5), continue; end
S(Sj)=median(single(S(convTri(single(Sj),1)>0))); del=1; end
if(del), [~,~,S]=unique(S); S=reshape(S-1,gtWidth,gtWidth);
segs(:,:,i)=S; nSegs(i)=max(S(:))+1; end
end
model.segs=segs; model.nSegs=nSegs;
% store compact representations of sparse binary edge patches
nBnds=opts.sharpen+1; eBins=cell(nTrees*nNodes,nBnds);
eBnds=zeros(nNodes*nTrees,nBnds);
parfor i=1:nTrees*nNodes
if(model.child(i) || model.nSegs(i)==1), continue; end %#ok<PFBNS>
E=gradientMag(single(model.segs(:,:,i)))>.01; E0=0;
% eBins stores the sparse edge coordinates for all segments
% eBnds stores the number of edge pixels for all segments
for j=1:nBnds, eBins{i,j}=uint16(find(E & ~E0)'-1); E0=E;
eBnds(i,j)=length(eBins{i,j}); E=convTri(single(E),1)>.01; end
end
eBins=eBins'; model.eBins=[eBins{:}]';
% eBnds now stores the index of each sparse edge structure
eBnds=eBnds'; model.eBnds=uint32([0; cumsum(eBnds(:))]);
end
%% the main function for training
function trainTree( trnImgDir, trnGtDir, opts, stream, treeInd )
% Train a single tree in forest model.
% location of ground truth
% note we will only train on the images with GT results
imgIds=dir(fullfile(trnGtDir, '*.png'));
fileExt = '.png';
if isempty(imgIds);
imgIds = dir(fullfile(trnGtDir, '*.jpg'));
fileExt = '.jpg';
end
imgIds=imgIds([imgIds.bytes]>0);
imgIds={imgIds.name};
nImgs=length(imgIds);
for i=1:nImgs,
imgIds{i}=imgIds{i}(1:end-4);
end
% extract commonly used options
imWidth=opts.imWidth; imRadius=imWidth/2;
gtWidth=opts.gtWidth; gtRadius=gtWidth/2;
nChns=opts.nChns; nTotFtrs=opts.nTotFtrs;
nPos=opts.nPos; nNeg=opts.nNeg; shrink=opts.shrink;
% finalize setup
treeDir = [opts.modelDir '/tree/'];
treeFn = [treeDir opts.modelFnm '_tree'];
if(exist([treeFn int2str2(treeInd,3) '.mat'],'file'))
fprintf('Reusing tree %d of %d\n',treeInd,opts.nTrees); return; end
fprintf('\n-------------------------------------------\n');
fprintf('Training tree %d of %d\n',treeInd,opts.nTrees); tStart=clock;
% set global stream to stream with given substream (will undo at end)
streamOrig = RandStream.getGlobalStream();
set(stream,'Substream',treeInd);
RandStream.setGlobalStream( stream );
% collect positive and negative patches and compute features
fids=sort(randperm(nTotFtrs,round(nTotFtrs*opts.fracFtrs)));
k = nPos+nNeg; nImgs=min(nImgs,opts.nImgs);
ftrs = zeros(k,length(fids),'single');
labels = zeros(gtWidth,gtWidth,k,'uint8'); k = 0;
tid = ticStatus('Collecting data',30,1);
% modified by YL
for i = 1:nImgs
% get image and compute channels
me=imread(fullfile(trnGtDir, [imgIds{i} fileExt]));
me = im2double(me);
I=imread(fullfile(trnImgDir, [imgIds{i} fileExt]));
nGt = opts.nGt;
% resize the image and groundtruth if necessary
if opts.scale < 1
%me = imresize(me, opts.scale);
I = imresize(I, opts.scale);
assert(size(I,1)==size(me,1) && size(I,2)==size(me,2))
end
% get image features
siz=size(I);
p=zeros(1,4); p([2 4])=mod(4-mod(siz(1:2),4),4);
if(any(p)), I=imPad(I,p,'symmetric'); end
[chnsReg,chnsSim] = edgesChns(I,opts);
% sample positive and negative locations
xy=[]; k1=0; B=false(siz(1),siz(2));
B(shrink:shrink:end,shrink:shrink:end)=1;
B([1:imRadius end-imRadius:end],:)=0;
B(:,[1:imRadius end-imRadius:end])=0;
% generate the threshlist
thresh = exp(linspace(log(opts.threshBracket(1)), log(opts.threshBracket(2)), nGt));
gt = cell([1 nGt]);
% all we care about is the pretty positive samples!
Mneg=~(bwdist(me>0.1)<gtRadius);
posLabels = [];
for j=1:nGt
% threshold the motion boundary to get the a boundary map
Mpos=(me>thresh(j)); gt{j} = Mpos;
% get the pos sample (if there is a boundary pixel within the patch)
Mpos(bwdist(Mpos)<gtRadius)=1;
[y,x]=find(Mpos.*B); k2=min(length(y),ceil(2*nPos/nImgs/nGt));
rp=randperm(length(y),k2); y=y(rp); x=x(rp);
xy=[xy; x y ones(k2,1)*j]; k1=k1+k2; %#ok<AGROW>
posLabels = [posLabels; ones(k2, 1)];
% lower thrshold for neg samples
[y,x]=find(Mneg.*B); k2=min(length(y),ceil(nNeg/nImgs/nGt));
rp=randperm(length(y),k2); y=y(rp); x=x(rp);
xy=[xy; x y ones(k2,1)*j]; k1=k1+k2; %#ok<AGROW>
posLabels = [posLabels; zeros(k2, 1)];
end
% k1 is the maximum number of samples, but we do not know the exact
% number until we sample them
k1 = ceil((nPos + nNeg)/nImgs); kSamples = length(posLabels);
if(k1>size(ftrs,1)-k), k1=size(ftrs,1)-k; xy=xy(1:k1,:); end
% crop patches and ground truth labels
psReg=zeros(imWidth/shrink,imWidth/shrink,nChns,k1,'single');
lbls=zeros(gtWidth,gtWidth,k1,'uint8');
psSim=psReg; ri=imRadius/shrink; rg=gtRadius;
% check each local patch
curSampleIdx = 0;
for j=1:kSamples,
if curSampleIdx < k1
% get the sample coordinate
xy1=xy(j,:); xy2=xy1/shrink;
% crop boundary patch -> find super pixel using boundary map
% -> remove boundary pixels
e = gt{xy1(3)}(xy1(2)-rg+1:xy1(2)+rg,xy1(1)-rg+1:xy1(1)+rg);
s = bwlabel(~e, 4);
t = spDetectMex('boundaries',uint32(s),single(e),0,1); t = t+ 1;
if(all(t(:)==t(1)) && posLabels(j)==1),
continue; % this is not a good edge sample, so skip it
elseif all(t(:)==t(1))
curSampleIdx = curSampleIdx + 1;
lbls(:,:,curSampleIdx)=1; % negtive sample
else
curSampleIdx = curSampleIdx + 1;
[~,~,t]=unique(t);
lbls(:,:,curSampleIdx)=reshape(t,gtWidth,gtWidth); % positive sample
end
% crop the feature
psReg(:,:,:,curSampleIdx)=chnsReg(xy2(2)-ri+1:xy2(2)+ri,xy2(1)-ri+1:xy2(1)+ri,:);
psSim(:,:,:,curSampleIdx)=chnsSim(xy2(2)-ri+1:xy2(2)+ri,xy2(1)-ri+1:xy2(1)+ri,:);
end
end
% the actual number of samples
k1 = curSampleIdx; psReg = psReg(:,:,:,1:k1); psSim = psSim(:,:,:,1:k1);
lbls = lbls(:,:,1:k1);
% visualization code
% if(1), figure(1); montage2(squeeze(psReg(:,:,1,:))); drawnow; end
% if(1), figure(2); montage2(lbls(:,:,:)); drawnow; end
% if(1), figure(3); imshow(me); drawnow; end
% pause
% compute features and store
ftrs1=[reshape(psReg,[],k1)' stComputeSimFtrs(psSim,opts)];
ftrs(k+1:k+k1,:)=ftrs1(:,fids); labels(:,:,k+1:k+k1)=lbls;
k=k+k1; if(k==size(ftrs,1)), tocStatus(tid,1); break; end
tocStatus(tid,i/nImgs);
end
if(k<size(ftrs,1)), ftrs=ftrs(1:k,:); labels=labels(:,:,1:k); end
% train structured edge classifier (random decision tree)
pTree=struct('minCount',opts.minCount, 'minChild',opts.minChild, ...
'maxDepth',opts.maxDepth, 'H',opts.nClasses, 'split',opts.split);
t=labels; labels=cell(k,1); for i=1:k, labels{i}=t(:,:,i); end
pTree.discretize=@(hs,H) discretize(hs,H,opts.nSamples,opts.discretize);
tree=forestTrain(ftrs,labels,pTree); tree.hs=cell2array(tree.hs);
tree.fids(tree.child>0) = fids(tree.fids(tree.child>0)+1)-1;
if(~exist(treeDir,'dir')), mkdir(treeDir); end
save([treeFn int2str2(treeInd,3) '.mat'],'tree'); e=etime(clock,tStart);
fprintf('Training of tree %d complete (time=%.1fs).\n',treeInd,e);
RandStream.setGlobalStream( streamOrig );
end
%% sim features
function ftrs = stComputeSimFtrs( chns, opts )
% Compute self-similarity features (order must be compatible w mex file).
w=opts.imWidth/opts.shrink; n=opts.nCells; if(n==0), ftrs=[]; return; end
nSimFtrs=opts.nSimFtrs; nChns=opts.nChns; m=size(chns,4);
inds=round(w/n/2); inds=round((1:n)*(w+2*inds-1)/(n+1)-inds+1);
chns=reshape(chns(inds,inds,:,:),n*n,nChns,m);
ftrs=zeros(nSimFtrs/nChns,nChns,m,'single');
k=0; for i=1:n*n-1, k1=n*n-i; i1=ones(1,k1)*i;
ftrs(k+1:k+k1,:,:)=chns(i1,:,:)-chns((1:k1)+i,:,:); k=k+k1; end
ftrs = reshape(ftrs,nSimFtrs,m)';
end
%% discretize the edge patches
function [hs,segs] = discretize( segs, nClasses, nSamples, type )
% Convert a set of segmentations into a set of labels in [1,nClasses].
persistent cache; w=size(segs{1},1); assert(size(segs{1},2)==w);
if(~isempty(cache) && cache{1}==w), [~,is1,is2]=deal(cache{:}); else
% compute all possible lookup inds for w x w patches
is=1:w^4; is1=floor((is-1)/w/w); is2=is-is1*w*w; is1=is1+1;
kp=is2>is1; is1=is1(kp); is2=is2(kp); cache={w,is1,is2};
end
% compute n binary codes zs of length nSamples
nSamples=min(nSamples,length(is1)); kp=randperm(length(is1),nSamples);
n=length(segs); is1=is1(kp); is2=is2(kp); zs=false(n,nSamples);
for i=1:n, zs(i,:)=segs{i}(is1)==segs{i}(is2); end
zs=bsxfun(@minus,zs,sum(zs,1)/n); zs=zs(:,any(zs,1));
if(isempty(zs)), hs=ones(n,1,'uint32'); segs=segs{1}; return; end
% find most representative segs (closest to mean)
[~,ind]=min(sum(zs.*zs,2)); segs=segs{ind};
% apply PCA to reduce dimensionality of zs
U=pca(zs'); d=min(5,size(U,2)); zs=zs*U(:,1:d);
% discretize zs by clustering or discretizing pca dimensions
d=min(d,floor(log2(nClasses))); hs=zeros(n,1);
for i=1:d, hs=hs+(zs(:,i)<0)*2^(i-1); end
[~,~,hs]=unique(hs); hs=uint32(hs);
if(strcmpi(type,'kmeans'))
nClasses1=max(hs); C=zs(1:nClasses1,:);
for i=1:nClasses1, C(i,:)=mean(zs(hs==i,:),1); end
hs=uint32(kmeans2(zs,nClasses,'C0',C,'nIter',1));
end
% optionally display different types of hs
for i=1:0, figure(i); montage2(cell2array(segs(hs==i))); end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
rgb.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/rgb.m
| 7,732 |
utf_8
|
e3e60031f5dfe6aa0d1f477e521ca66d
|
% rgb.m: translates a colour from multiple formats into matlab colour format
% type 'rgb demo' to get started
%
% [matlabcolor]=rgb(col)
% matlab colors are in the format [R G B]
%
% if 'col' is a string, it is interpreted as
%
% [[modifier] descriptor] colour_name
%
% where
% modifier is one of (slightly, normal, very, extremely)
% descriptor is one of (light/pale, normal, dark)
% colorname is a name of a colour
% (type 'rgb list' or 'rgb demo' to see them all)
%
% if 'col' is an integer between 0 and &HFFFFFF inclusive,
% it is interpreted as a double word RGB value in the form
% [0][R][G][B]
%
% if 'col' is a negative integer between -1 and -&HFFFFFF
% inclusive, it is interpreted as the complement of a double
% word RGB value in the form [0][B][G][R]
%
% if 'col' is a string of the form 'qbX' or 'qbXX' where X
% is a digit then the number part is interpreted as a qbasic
% color
%
% if 'col' is one of {k,w,r,g,b,y,m,c} a sensible result is
% returned
%
% if 'col' is already in matlab format, it is unchanged
% VERSION: 06/06/2002
% AUTHOR: ben mitch
% CONTACT: [email protected]
% WWW: www.benmitch.co.uk\matlab (not yet)
% LOCATION: figures\colors\
function out=rgb(in)
if isa(in,'char') & length(in)>2 & length(in)<5 & strcmpi('qb',in(1:2))
out=qbcolor(sscanf(in(3:end),'%i'));
elseif isa(in,'char') & length(in)==1
out=translatecolorchar(in);
elseif isa(in,'char')
if strcmp(in,'demo') rgb_demo; return; end
if strcmp(in,'list') rgb_list; return; end
out=translatecolorstring(in);
elseif isa(in,'double') & size(in,1)==1 & size(in,2)==1 & abs(in)<16777216
out=translatecolorRGB(in);
elseif isa(in,'double') & size(in,1)==1 & size(in,2)==3
out=in;
else
warning('Unrecognised color format, black assumed');
out=[0 0 0];
end
function out=translatecolorchar(in)
switch(in)
case 'k', out=[0 0 0];
case 'w', out=[1 1 1];
case 'r', out=[1 0 0];
case 'g', out=[0 1 0];
case 'b', out=[0 0 1];
case 'y', out=[1 1 0];
case 'm', out=[1 0 1];
case 'c', out=[0 1 1];
otherwise
warning(['Unrecognised colour "' in '", black assumed'])
out=[0 0 0];
return;
end
function out=translatecolorstring(in)
args.tokens=rgb_parse(in);
args.N=length(args.tokens);
if args.N>3 warning('Too many words in color description, any more than 3 will be ignored'); end
while(args.N<3)
args.tokens=[{'normal'};args.tokens];
args.N=args.N+1;
end
cols=get_cols;
col=[];
for n=1:size(cols,1)
names=cols{n,1};
for m=1:length(names)
if strcmp(args.tokens{3},names{m}) col=cols{n,2}; break; end
end
if ~isempty(col) break; end
end
if isempty(col)
warning(['Unrecognised colour "' args.tokens{3} '", black assumed'])
out=[0 0 0];
return;
end
switch args.tokens{1}
case 'slightly', fac=0.75;
case 'normal', fac=0.5;
case 'very', fac=0.25;
case 'extremely', fac=0.125;
otherwise
warning(['Unrecognised modifier "' args.tokens{1} '", normal assumed'])
fac=0.5;
end
switch args.tokens{2}
case {'light','pale'}, out=1-(1-col)*fac;
case 'normal', out=col;
case 'dark', out=col*fac;
otherwise
warning(['Unrecognised descriptor "' args.tokens{2} '", normal assumed'])
out=col;
end
function out=translatecolorRGB(in)
BGR=0;
if in<0
in=-in;
BGR=1;
end
b=bytes4(in);
if BGR out=b(4:-1:2); else out=b(2:4); end
function out=qbcolor(in)
% rgb value from basic colour code
% 0-7 are normal, 8-15 are bright
% 0 - black
% 1 - red, 2 - green, 3 - blue
% 4 - cyan, 5 - magenta, 6 - yellow
% 7 - white
bright=0.5;
if in>7 in=in-8; bright=1; end
switch in
case 0, rgb=[0 0 0];
case 1, rgb=[1 0 0];
case 2, rgb=[0 1 0];
case 3, rgb=[0 0 1];
case 4, rgb=[0 1 1];
case 5, rgb=[1 0 1];
case 6, rgb=[1 1 0];
case 7, rgb=[1 1 1];
otherwise
warning('Unrecognised QBasic color, black assumed');
out=[0 0 0];
return;
end
out=rgb*bright;
function tokens=rgb_parse(str)
% parse string to obtain all tokens
% quoted strings count as single tokens
inquotes=0;
intoken=0;
pos=1;
l=length(str);
st=0;
ed=0;
token='';
tab=char(9);
tokens=cell(0);
while(pos<=l)
ch=str(pos);
if inquotes
if ch=='"'
inquotes=0;
tokens={tokens{:} token};
else
token=[token ch];
end
elseif intoken
if ch==' ' | ch==tab
intoken=0;
tokens={tokens{:} token};
elseif ch=='"'
error(['Quote misplace in <' str '>']);
else
token=[token ch];
end
else
if ch==' ' | ch==tab
% do nothing
elseif ch=='"'
token='';
inquotes=1;
else
token=ch;
intoken=1;
end
end
pos=pos+1;
end
if intoken tokens={tokens{:} token}; end
if inquotes error(['Unpaired quotes in <' str '>']); end
tokens=tokens';
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% DEMO
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function rgb_demo
figure(1)
clf
cols = get_cols;
cols = {cols{:,1}}';
cols = { cols{:}, ...
'k', ...
'r', ...
'g', ...
'b', ...
'y', ...
'm', ...
'c', ...
'w', ...
'', ...
'extremely dark green', ...
'very dark green', ...
'dark green', ...
'slightly dark green', ...
'green', ...
'slightly pale green', ...
'pale green', ...
'very pale green', ...
'extremely pale green', ...
};
height=9;
x=0;
y=0;
for n=1:length(cols)
rect(x,y,cols{n})
y=y+1;
if y==height
x=x+2;
y=0;
end
end
if y==0 x=x-2; end
axis([0 (x+2) 0 height])
title('names on different rows are alternates')
function rect(x,y,col)
if isempty(col) return; end
r=rectangle('position',[x+0.1 y+0.1 1.8 0.8]);
col_=col;
if iscell(col) col=col{1}; end
colrgb=rgb(col);
if strcmp(col(1),'u') & length(col)==2
t=text(x+1,y+0.5,{'unnamed',['colour (' col(2) ')']});
set(r,'facecolor',colrgb);
else
t=text(x+1,y+0.5,col_);
set(r,'facecolor',colrgb);
if sum(colrgb)<1.5 set(t,'color',[1 1 1]); end
end
set(t,'horizontalalignment','center')
set(t,'fontsize',10)
function rgb_list
cols=get_cols;
disp(' ')
for n=1:size(cols,1)
code=cols{n,2};
str=cols{n,1};
str_=[];
for m=1:length(str)
str_=[str_ str{m} ', '];
end
str_=str_(1:end-2);
if strcmp(str_(1),'u') & length(str_)==2
str_=['* (' str_(2) ')'];
end
disp([' [' sprintf('%.1f %.1f %.1f',code) '] - ' str_])
end
disp([10 '* colours marked thus are not named - if you know their' 10 ' designation, or if you feel sure a colour is mis-named,' 10 ' email me (address via help) or comment at' 10 ' www.mathworks.com/matlabcentral - "rgb demo" to see them' 10])
function cols=get_cols
cols={
'black', [0 0 0]; ...
'navy', [0 0 0.5]; ...
'blue', [0 0 1]; ...
'u1', [0 0.5 0]; ...
{'teal','turquoise'}, [0 0.5 0.5]; ...
'slateblue', [0 0.5 1]; ...
{'green','lime'}, [0 1 0]; ...
'springgreen', [0 1 0.5]; ...
{'cyan','aqua'}, [0 1 1]; ...
'maroon', [0.5 0 0]; ...
'purple', [0.5 0 0.5]; ...
'u2', [0.5 0 1]; ...
'olive', [0.5 0.5 0]; ...
{'gray','grey'}, [0.5 0.5 0.5]; ...
'u3', [0.5 0.5 1]; ...
{'mediumspringgreen','chartreuse'}, [0.5 1 0]; ...
'u4', [0.5 1 0.5]; ...
'sky', [0.5 1 1]; ...
'red', [1 0 0]; ...
'u5', [1 0 0.5]; ...
{'magenta','fuchsia'}, [1 0 1]; ...
'orange', [1 0.5 0]; ...
'u6', [1 0.5 0.5]; ...
'u7', [1 0.5 1]; ...
'yellow', [1 1 0]; ...
'u8', [1 1 0.5]; ...
'white', [1 1 1]; ...
};
for n=1:size(cols,1)
if ~iscell(cols{n,1}) cols{n,1}={cols{n,1}}; end
end
% converts a DWORD into a four byte row vector
function out=bytes4(in)
out=[0 0 0 0];
if in<0 | in>(2^32-1)
warning('DWORD out of range, zero assumed');
return;
end
N=4;
while(in>0)
out(N)=mod(in,256);
in=(in-out(N))/256;
N=N-1;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
newMPPObject.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/newMPPObject.m
| 601 |
utf_8
|
42d488d286b1fadfe14e55d2aea0209a
|
% --
% creates a template for MotionplayerPro Objects
% fields required:
% - type (currently possible values: 'dot', 'cross', 'tetra'
% - data: matrix of size (3*nrOfObj x nrOfFrames)
% - samplingRate
% optional fields: (for default values see DEFAULTSCENE)
% - color
% - alpha
% - size
%
% author: Jochen Tautges ([email protected])
function object = newMPPObject()
% help newMPPObject;
object.type = [];
object.data = [];
object.samplingRate = [];
object.color = [];
object.alpha = [];
object.size = [];
object.boundingBox = [];
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
MPP_GUI.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/MPP_GUI.m
| 3,405 |
utf_8
|
aedd8a008aa2e4af018cdaa12503396a
|
function fig = MPP_GUI(varargin)
global SCENE;
% figure and camera settings ----------------------------------------------
fig = figure('Visible','on',...
'Name','MotionPlayerPro',...
'NumberTitle','off',...
'Position',[SCENE.position,SCENE.size],...
'Resize', 'on', ...
'Color', [0 0 0 ], ...%[.92 .95 .95],...
'Interruptible','on',...
'WindowScrollWheelFcn',@windowScrollWheelFcn,...
'KeyPressFcn',@keyPressFunction,...
'KeyReleaseFcn',@keyReleaseFunction,...
'Renderer','OpenGl');
% 'WindowButtonDownFcn',@mouseButtonDownFunction,...
% 'WindowButtonUpFcn',@mouseButtonUpFunction,...
SCENE.handles.light = light('Position',SCENE.lightPosition,'Style','infinite');
set(SCENE.handles.light,'Visible',SCENE.status.light);
SCENE.keyEvents.shiftKeyDown = false;
SCENE.keyEvents.altKeyDown = false;
cameratoolbar(fig, 'Show');
cameratoolbar(fig, 'SetCoordSys',SCENE.status.mainAxis);
axis equal;
axis off;
hold all;
function keyPressFunction(src,evnt)
switch(evnt.Key)
case 'leftarrow'
curFrame = max(SCENE.status.curFrame - 10,1);
setFramePro(curFrame);
drawnow();
case 'downarrow'
curFrame = max(SCENE.status.curFrame - 100,1);
setFramePro(curFrame);
drawnow();
case 'rightarrow'
curFrame = min(SCENE.status.curFrame + 10,SCENE.nframes);
setFramePro(curFrame);
drawnow();
case 'uparrow'
curFrame = min(SCENE.status.curFrame + 100,SCENE.nframes);
setFramePro(curFrame);
drawnow();
case 'shift'
if(~SCENE.keyEvents.shiftKeyDown)
SCENE.keyEvents.shiftKeyDown = true;
cameratoolbar(SCENE.handles.fig, 'SetCoordSys',SCENE.status.mainAxis);
cameratoolbar(SCENE.handles.fig, 'SetMode','orbit');
% set(cam_Status_Label,'String','orbit');
end
case 'alt'
if(~SCENE.keyEvents.altKeyDown)
SCENE.keyEvents.altKeyDown = true;
cameratoolbar(SCENE.handles.fig, 'SetCoordSys',SCENE.status.mainAxis);
cameratoolbar(SCENE.handles.fig, 'SetMode','pan');
% set(cam_Status_Label,'String','pan');
end
case 'space'
if(SCENE.status.running)
pauseFunction;
else
playFunction;
end
otherwise
disp('unknown key');
end
end
function keyReleaseFunction(src,evnt)
switch(evnt.Key)
case 'shift'
SCENE.keyEvents.shiftKeyDown = false;
cameratoolbar(SCENE.handles.fig, 'SetCoordSys','none');
cameratoolbar(SCENE.handles.fig, 'SetMode','nomode');
case 'alt'
SCENE.keyEvents.altKeyDown = false;
cameratoolbar(SCENE.handles.fig, 'SetCoordSys','none');
cameratoolbar(SCENE.handles.fig, 'SetMode','nomode');
end
end
function windowScrollWheelFcn(src, evnt)
f = .05;
if(evnt.VerticalScrollCount < 0)
zoom(1+f);
else
zoom(1-f);
end
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
motionplayerProGUI.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/motionplayerProGUI.m
| 29,219 |
utf_8
|
ae6f6ca6e5c3b23a95a5b71b80b6d6be
|
function fig = motionplayerProGUI(varargin)
global SCENE;
% help --------------------------------------------------------------------
helpDlg = {'Key-Bindings:',...
'------------------------------------------------------',...
'<space>:',' play/pause',...
'<leftarrow>:',' move motion back 10 frames',...
'<downarrow>:',' move motion back 100 frames',...
'<rightarrow>:',' move motion forward 10 frames',...
'<uparrow>:',' move motion forward 100 frames',...
'<shift>+<mouse>:',' orbit rotate scene',...
'<alt>+<mouse>:',' pan scene',...
'<mousewheel>:',' zoom scene'};
% figure and camera settings ----------------------------------------------
fig = figure('Visible','on',...
'Name','MotionPlayerPro',...
'NumberTitle','off',...
'Position',[SCENE.position,SCENE.size],...
'Resize', 'on', ...
'Color', [1 1 1], ...%[.92 .95 .95],...
'Interruptible','on',...
'WindowScrollWheelFcn',@windowScrollWheelFcn,...
'KeyPressFcn',@keyPressFunction,...
'KeyReleaseFcn',@keyReleaseFunction,...
'Renderer','OpenGl');
% 'WindowButtonDownFcn',@mouseButtonDownFunction,...
% 'WindowButtonUpFcn',@mouseButtonUpFunction,...
SCENE.handles.light = light('Position',SCENE.lightPosition,'Style','infinite');
set(SCENE.handles.light,'Visible',SCENE.status.light);
SCENE.keyEvents.shiftKeyDown = false;
SCENE.keyEvents.altKeyDown = false;
cameratoolbar(fig, 'Show');
cameratoolbar(fig, 'SetCoordSys',SCENE.status.mainAxis);
axis equal;
% axis vis3d;
axis off;
renewAxisDimensions(SCENE.boundingBox);
% initializing figure with first frame(s) ---------------------------------
for j=1:SCENE.nmots
if SCENE.nmots>1 % the interpolate skeleton colors
c = (j-1)/(SCENE.nmots-1);
interpolated_color = ...
c * SCENE.colors.multipleSkels_FaceVertexData_end...
+ (1-c) * SCENE.colors.multipleSkels_FaceVertexData_start;
end
nrOfNodes = size(SCENE.mots{j}.vertices{1},1)/3;
for i=1:size(SCENE.mots{j}.vertices,1)-1
v = reshape(SCENE.mots{j}.vertices{i+1}(:,1),3,nrOfNodes)';
f = SCENE.mots{j}.faces;
if SCENE.nmots==1
SCENE.mots{j}.joint_handles(i) = ...
patch('Vertices',v,'Faces',f,...
'FaceVertexCData',SCENE.colors.singleSkel_FaceVertexData,...
'FaceColor',SCENE.colors.singleSkel_FaceColor,...
'FaceAlpha',SCENE.colors.singelSkel_FaceAlpha, ...
'EdgeColor',SCENE.colors.singleSkel_EdgeColor);
else
SCENE.mots{j}.joint_handles(i) = ...
patch('Vertices',v,'Faces',f,...
'FaceColor',hsv2rgb(interpolated_color),...
'FaceAlpha',SCENE.colors.multipleSkels_FaceAlpha, ...
'EdgeColor',SCENE.colors.multipleSkels_EdgeColor);
end
end
SCENE.mots{j}.jointID_handles = -ones(SCENE.mots{j}.njoints,1);
end
hold all;
% % for j=1:SCENE.npoints
% % if SCENE.npoints>1
% % c = (j-1)/(SCENE.npoints-1);
% % else
% % c = 0;
% % end
% % color = ...
% % c * SCENE.colors.points_end...
% % + (1-c) * SCENE.colors.points_start;
% %
% % i = mod(j,numel(SCENE.pointsSpec));
% % if i==0, i=numel(SCENE.pointsSpec); end
% % SCENE.handles.points(j) = plot3(SCENE.points{j}(1:3:end,1),...
% % SCENE.points{j}(2:3:end,1),...
% % SCENE.points{j}(3:3:end,1),...
% % SCENE.pointsSpec{i},'color',hsv2rgb(color));
% % end
if ~isempty(SCENE.objects)
objects = fieldnames(SCENE.objects);
nrOfDiffObjects = numel(objects);
for p=1:nrOfDiffObjects
for k=1:SCENE.objects.(objects{p}).counter
color = SCENE.objects.(objects{p}).color{k};
switch objects{p}
case 'tetra'
coords = [1 -1 -1 1;1 -1 1 -1;-1 -1 1 1];
f = [1 2 3;1 2 4;1 3 4;2 3 4];
nrOfObj = size(SCENE.objects.tetra.procdata{k},1);
if ~isempty(SCENE.objects.tetra.alpha{k})
alphaValues = SCENE.objects.tetra.alpha{k}(:,1);
else
alphaValues = ones(nrOfObj,1);
end
SCENE.handles.tetra{k}=zeros(nrOfObj,1);
for n=1:nrOfObj
v = reshape(SCENE.objects.tetra.procdata{k}{n}(:,1),3,size(coords,2));
SCENE.handles.tetra{k}(n) = patch('Vertices',v','Faces',f,...
'FaceColor',color,...
'FaceAlpha',alphaValues(n), ...
'EdgeColor','none');
end
case {'dot','cross','circle'}
switch objects{p}
case 'dot', lineSpec = '.';
case 'cross', lineSpec = 'x';
case 'circle', lineSpec = 'o';
otherwise
error('Unknown obj type');
end
SCENE.handles.(objects{p}){k} = plot3(SCENE.objects.(objects{p}).procdata{k}(1:3:end,1),...
SCENE.objects.(objects{p}).procdata{k}(2:3:end,1),...
SCENE.objects.(objects{p}).procdata{k}(3:3:end,1),...
lineSpec,'color',color);
case {'line'}
nrOfLines = size(SCENE.objects.line.procdata{k},1);
nrOfJoints = size(SCENE.objects.line.procdata{k}{1},1)/3;
n=0;
color = rgb(SCENE.objects.line.color{k});
for n1=1:nrOfLines
for n2=1:nrOfJoints
n=n+1;
if ~isempty(SCENE.objects.line.alpha{k})
w = SCENE.objects.line.alpha{k}(n1,1);
color = w * color + (1-w) * [1 1 1];
end
SCENE.handles.line{k}(n) = plot3(SCENE.objects.line.procdata{k}{n1,1}(3*n2-2,:),...
SCENE.objects.(objects{p}).procdata{k}{n1,1}(3*n2-1,:),...
SCENE.objects.(objects{p}).procdata{k}{n1,1}(3*n2,:),...
'-','color',color);
end
end
end
end
end
end
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
hold off;
% if SCENE.nmots>1
% spreadFunction();
% end
%% control panel-----------------------------------------------------------
SCENE.handles.control_Panel = uipanel(...
'Parent',fig,...
'Units','pixels',...
'Position',[2 2 799 110],...
'BackgroundColor',[.97 .97 .97]);
SCENE.handles.goto_First_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.goto_First,...'String','|<',...
'Units','pixels',...
'Position',[2 80 27 20],...
'TooltipString','go to first frame',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@gotoFirstFunction);
SCENE.handles.play_reverse_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.play_reverse,...'String','<|',...
'Units','pixels',...
'Position',[30 80 27 20],...
'TooltipString','play backwards',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@playReverseFunction);
SCENE.handles.pause_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.pause,...'String','||',...
'Units','pixels',...
'Position',[58 80 27 20],...
'TooltipString','pause',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@pauseFunction);
SCENE.handles.play_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.play,...'String','|>',...
'Units','pixels',...
'Position',[86 80 27 20],...
'TooltipString','play',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@playFunction);
SCENE.handles.goto_Last_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.goto_Last,...'String','>|',...
'Units','pixels',...
'Position',[114 80 27 20],...
'TooltipString','go to last frame',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@gotoLastFunction);
SCENE.handles.slower_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.slower,...'String','<<',...
'Units','pixels',...
'Position',[148 80 27 20],...
'TooltipString','slower',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@slowerFunction);
SCENE.handles.loop_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.unlooped,...'String','|--|',...
'Units','pixels',...
'Position',[176 80 27 20],...
'TooltipString','loop',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@loopFunction);
SCENE.handles.faster_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.faster,...'String','>>',...
'Units','pixels',...
'Position',[204 80 27 20],...
'TooltipString','faster',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@fasterFunction);
SCENE.handles.drawCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.coords+0.5,...
'Units','pixels',...
'Position',[236 80 27 20],...
'TooltipString','draw coordinate system',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawCoordinateSystem);
SCENE.handles.drawLocalCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.localcoords+0.5,...
'Units','pixels',...
'Position',[264 80 27 20],...
'TooltipString','draw local coordinate systems',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawLocalCoordinateSystems2);
SCENE.handles.drawGroundPlane_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.groundPlane,...
'Units','pixels',...
'Position',[292 80 27 20],...
'TooltipString','hide ground plane',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawGroundPlane);
SCENE.handles.drawJointIDs_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.jointIDs+0.5,...
'Units','pixels',...
'Position',[320 80 27 20],...
'TooltipString','show joint IDs',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawJointIDs);
SCENE.handles.drawSensorCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.sensorcoords+0.5,...
'Units','pixels',...
'Position',[348 80 27 20],...
'TooltipString','draw sensor coordinate systems',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawSensorCoordinateSystems);
% SCENE.handles.drawSensorCoordSyst2_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'CData',SCENE.buttons.sensorcoords+0.5,...
% 'Units','pixels',...
% 'Position',[348 80 27 20],...
% 'TooltipString','draw sensor coordinate systems',...
% 'BackgroundColor',SCENE.colors.buttons_group2, ...
% 'CallBack',@drawSensorCoordinateSystems2);
if SCENE.nmots>1
SCENE.handles.spread_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.spread+0.5,...
'Units','pixels',...
'Position',[376 80 27 20],...
'TooltipString','spread motions',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@spreadFunction);
end
SCENE.handles.render_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.renderScene,...
'Units','pixels',...
'Position',[SCENE.size(1)-41-2*32 7 30 20],...
'TooltipString','render Scene to avi',...
'BackgroundColor',[0.9 0.3 0], ...
'CallBack',@renderMPProScene);
SCENE.handles.help_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'String','Help',...
'Units','pixels',...
'FontWeight','bold',...
'HorizontalAlignment','center',...
'Position',[SCENE.size(1)-41-32 6 30 22],...
'CallBack',@helpButtonFunction);
SCENE.handles.quit_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.quit,...'String','Quit',...
'Units','pixels',...
'Position',[SCENE.size(1)-41 7 30 20],...
'TooltipString','quit',...
'BackgroundColor',[0.9,.0,.0], ...
'CallBack',@closeFunction);
% SCENE.handles.axis_x_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','x',...
% 'Units','pixels',...
% 'Position',[284 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to x',...
% 'BackgroundColor',[0.8,0.8,0.8], ...
% 'CallBack',@setMainAxisFunction);
%
% SCENE.handles.axis_y_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','y',...
% 'Units','pixels',...
% 'Position',[306 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to y',...
% 'BackgroundColor',[0.9,0.9,0.97], ...
% 'CallBack',@setMainAxisFunction);
%
% SCENE.handles.axis_z_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','z',...
% 'Units','pixels',...
% 'Position',[328 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to z',...
% 'BackgroundColor',[0.8,0.8,0.8], ...
% 'CallBack',@setMainAxisFunction);
SCENE.handles.status_Panel = uipanel(...
'Parent',SCENE.handles.control_Panel,'Units','pixels',...
'Position',[2 2 SCENE.size(1)-8 30],...
'BackgroundColor',[.97 .97 .97]);
SCENE.handles.curFrameLabel = uicontrol(SCENE.handles.status_Panel,'Style','Text', ...
'String',sprintf(' 1 / %d (%.2f s)', SCENE.nframes,0),...
'Units','pixels',...
'TooltipString','current frame',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[1 0 130 22]);
SCENE.handles.curSpeedLabel = uicontrol(SCENE.handles.status_Panel,'Style','Text', ...
'String','x 1.000',...
'Units','pixels',...
'TooltipString','current speed',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[150 0 40 22]);
% add frame markers above slider ------------------------------------------
if(SCENE.nframes > 1)
SCENE.handles.sliderHandle = uicontrol(SCENE.handles.control_Panel,'Style','Slider', ...
'String','Current Frame',...
'Units','pixels',...
'Max',SCENE.nframes,...
'Min',1,...
'Value',1,...
'SliderStep',[1/SCENE.nframes (1/SCENE.size(1))*40],...
'Position',[2 35 SCENE.size(1)-8 20],...
'BackgroundColor',[.8 .8 .8], ...
'CallBack',@moveFrameSliderFunction);
if SCENE.nframes<=20
numMarks = SCENE.nframes;
else
numMarks = 15;
end
for i=20:-1:5
if mod(SCENE.nframes-1,i)==0
numMarks=i+1;
break;
end
end
posFromLeft = 11;
posFromRight = SCENE.size(1)-62;
posFromLeft = posFromLeft-(posFromRight-posFromLeft)/(SCENE.nframes-1);
for frameNum = 1:(SCENE.nframes-1)/(numMarks-1):SCENE.nframes
uicontrol(SCENE.handles.control_Panel,'Style','Text',...
'String',round(frameNum),'Units','pixels',...
'FontSize',7,'BackgroundColor',[.97 .97 .97],...
'Position',[posFromLeft+(round(frameNum)/SCENE.nframes)*(posFromRight-posFromLeft) 60 45 12]);
end
end
if SCENE.status.spread
for n=1:SCENE.nmots
if SCENE.mots{n}.rotDataAvailable
spreadVertices(n);
else
fprintf('Note: Transformation of point clouds (c3d) is not yet supported!\n');
end
end
computeBoundingBoxSCENE();
set(SCENE.handles.spread_Button, 'CData',SCENE.buttons.spread,'TooltipString','unspread motions');
SCENE.status.spread = true;
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
renewAxisDimensions(SCENE.boundingBox);
setFramePro(SCENE.status.curFrame);
end
%% callback functions -----------------------------------------------------
function playReverseFunction(varargin)
if (SCENE.status.curFrame == 1)
SCENE.status.curFrame = SCENE.nframes;
end
SCENE.status.reverse = true;
SCENE.status.running = true;
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
while SCENE.status.running && (SCENE.status.curFrame>1 || SCENE.status.looped)
nextFrame = getNextFrame_local();
if SCENE.status.running
setFramePro(nextFrame);
drawnow;
SCENE.status.timeStamp = SCENE.timeOffset-toc*SCENE.status.speed;
end
end
SCENE.status.running = false;
end
function pauseFunction(varargin)
SCENE.status.running = false;
end
function playFunction(varargin)
if (SCENE.status.curFrame == SCENE.nframes)
SCENE.status.curFrame = 1;
end
SCENE.status.reverse = false;
SCENE.status.running = true;
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
while SCENE.status.running && (SCENE.status.curFrame<SCENE.nframes || SCENE.status.looped)
nextFrame = getNextFrame_local();
if SCENE.status.running
setFramePro(nextFrame);
drawnow;
SCENE.status.timeStamp = SCENE.timeOffset+toc*SCENE.status.speed;
end
end
SCENE.status.running = false;
end
function gotoFirstFunction(varargin)
SCENE.status.running = false;
SCENE.status.curFrame = 1;
setFramePro(1);
drawnow();
end
function gotoLastFunction(varargin)
SCENE.status.running = false;
SCENE.status.curFrame = SCENE.nframes;
setFramePro(SCENE.nframes);
drawnow();
end
function closeFunction(varargin)
SCENE.status.running = false;
close;
SCENE.objects = [];
SCENE.mots = [];
SCENE.nmots = 0;
SCENE.skels = [];
SCENE.nskels = 0;
% clear global SCENE;
end
function slowerFunction(varargin)
if (SCENE.status.speed > 0.125)
SCENE.status.speed = SCENE.status.speed/2;
set(SCENE.handles.curSpeedLabel,'String',...
sprintf('x %1.3f',SCENE.status.speed));
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
end
end
function fasterFunction(varargin)
if (SCENE.status.speed < 8)
SCENE.status.speed = SCENE.status.speed*2;
set(SCENE.handles.curSpeedLabel,'String',...
sprintf('x %1.3f',SCENE.status.speed));
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
end
end
function loopFunction(varargin)
if (SCENE.status.looped)
SCENE.status.looped = false;
set(SCENE.handles.loop_Button, 'CData',SCENE.buttons.unlooped,'TooltipString','loop');
else
SCENE.status.looped = true;
set(SCENE.handles.loop_Button, 'CData',SCENE.buttons.looped,'TooltipString','no loop');
end
end
function drawJointIDs(varargin)
if SCENE.status.jointIDs_drawn
SCENE.status.jointIDs_drawn = false;
for ii=1:SCENE.nmots
arrayfun(@(x) delete(x), SCENE.mots{ii}.jointID_handles);
end
set(SCENE.handles.drawJointIDs_Button,'CData',SCENE.buttons.jointIDs+0.5,'TooltipString','show joint IDs');
else
SCENE.status.jointIDs_drawn = true;
if ~SCENE.status.running
setFramePro(SCENE.status.curFrame);
end
set(SCENE.handles.drawJointIDs_Button,'CData',SCENE.buttons.jointIDs,'TooltipString','hide joint IDs');
end
end
function moveFrameSliderFunction(varargin)
if (SCENE.status.running)
SCENE.status.running = false;
end
curFrame = round(get(SCENE.handles.sliderHandle,'Value'));
SCENE.status.timeStamp = curFrame/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
setFramePro(curFrame);
drawnow;
end
function spreadFunction(varargin)
if SCENE.status.spread == false
for m=1:SCENE.nmots
if SCENE.mots{m}.rotDataAvailable
spreadVertices(m);
else
fprintf('Note: Transformation of point clouds (c3d) is not yet supported!\n');
end
end
computeBoundingBoxSCENE();
set(SCENE.handles.spread_Button, 'CData',SCENE.buttons.spread,'TooltipString','unspread motions');
SCENE.status.spread = true;
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
renewAxisDimensions(SCENE.boundingBox);
else
for m=1:SCENE.nmots
if SCENE.mots{m}.rotDataAvailable
unspreadVertices(m);
end
end
computeBoundingBoxSCENE();
set(SCENE.handles.spread_Button, 'CData',SCENE.buttons.spread+0.5,'TooltipString','spread motions');
SCENE.status.spread = false;
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
renewAxisDimensions(SCENE.boundingBox);
end
setFramePro(SCENE.status.curFrame);
end
% function setMainAxisFunction(varargin)
% buttonHandle = get(varargin{1,1});
% mA = buttonHandle.String;
% switch(mA)
% case 'x'
% SCENE.status.mainAxis = 'x';
% set(SCENE.handles.axis_x_Button,'BackgroundColor',[.9 .9 .97]);
% set(SCENE.handles.axis_y_Button,'BackgroundColor',[.8 .8 .8]);
% set(SCENE.handles.axis_z_Button,'BackgroundColor',[.8 .8 .8]);
% case 'y'
% SCENE.status.mainAxis = 'y';
% set(SCENE.handles.axis_x_Button,'BackgroundColor',[.8 .8 .8]);
% set(SCENE.handles.axis_y_Button,'BackgroundColor',[.9 .9 .97]);
% set(SCENE.handles.axis_z_Button,'BackgroundColor',[.8 .8 .8]);
% case 'z'
% SCENE.status.mainAxis = 'z';
% set(SCENE.handles.axis_x_Button,'BackgroundColor',[.8 .8 .8]);
% set(SCENE.handles.axis_y_Button,'BackgroundColor',[.8 .8 .8]);
% set(SCENE.handles.axis_z_Button,'BackgroundColor',[.9 .9 .97]);
% end
% cameratoolbar(fig, 'SetCoordSys',SCENE.status.mainAxis);
% end
function drawGroundPlane(varargin)
if SCENE.status.groundPlane_drawn
SCENE.status.groundPlane_drawn = false;
% set(SCENE.handles.groundPlane,'Visible','off');
set(SCENE.handles.drawGroundPlane_Button,'CData',SCENE.buttons.groundPlane+0.5,'TooltipString','draw ground plane');
set(SCENE.handles.groundPlane,'FaceAlpha',0,'EdgeColor','none');
else
SCENE.status.groundPlane_drawn = true;
% set(SCENE.handles.groundPlane,'Visible','on');
set(SCENE.handles.drawGroundPlane_Button,'CData',SCENE.buttons.groundPlane,'TooltipString','hide ground plane');
set(SCENE.handles.groundPlane,'FaceAlpha',0.7,'EdgeColor','black');
end
end
function keyPressFunction(src,evnt)
switch(evnt.Key)
case 'leftarrow'
curFrame = max(SCENE.status.curFrame - 10,1);
setFramePro(curFrame);
drawnow();
case 'downarrow'
curFrame = max(SCENE.status.curFrame - 100,1);
setFramePro(curFrame);
drawnow();
case 'rightarrow'
curFrame = min(SCENE.status.curFrame + 10,SCENE.nframes);
setFramePro(curFrame);
drawnow();
case 'uparrow'
curFrame = min(SCENE.status.curFrame + 100,SCENE.nframes);
setFramePro(curFrame);
drawnow();
case 'shift'
if(~SCENE.keyEvents.shiftKeyDown)
SCENE.keyEvents.shiftKeyDown = true;
cameratoolbar(fig, 'SetCoordSys',SCENE.status.mainAxis);
cameratoolbar(fig, 'SetMode','orbit');
% set(cam_Status_Label,'String','orbit');
end
case 'alt'
if(~SCENE.keyEvents.altKeyDown)
SCENE.keyEvents.altKeyDown = true;
cameratoolbar(fig, 'SetCoordSys',SCENE.status.mainAxis);
cameratoolbar(fig, 'SetMode','pan');
% set(cam_Status_Label,'String','pan');
end
case 'space'
if(SCENE.status.running)
pauseFunction;
else
playFunction;
end
otherwise
disp('unknown key');
end
end
function keyReleaseFunction(src,evnt)
switch(evnt.Key)
case 'shift'
SCENE.keyEvents.shiftKeyDown = false;
cameratoolbar(fig, 'SetCoordSys','none');
cameratoolbar(fig, 'SetMode','nomode');
case 'alt'
SCENE.keyEvents.altKeyDown = false;
cameratoolbar(fig, 'SetCoordSys','none');
cameratoolbar(fig, 'SetMode','nomode');
end
end
function windowScrollWheelFcn(src, evnt)
f = .05;
if(evnt.VerticalScrollCount < 0)
zoom(1+f);
else
zoom(1-f);
end
end
function helpButtonFunction(src,evnt)
% msgbox(helpDlg,'Help','help');
helpFig = figure( 'Visible','on',...
'Name','Help',...
'NumberTitle','off',...
'Menu','none',...
'Position',[400,200,250,400],...
'Resize', 'on', ...
'Color',[.92 .95 .95]);
uicontrol(helpFig,'Style','Text', ...
'String',helpDlg,...
'Units','pixels',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[0 0 250 400]);
end
end
%% local functions --------------------------------------------------------
function nextFrame = getNextFrame_local()
global SCENE;
framesToDrop = SCENE.status.timeStamp*SCENE.samplingRate-SCENE.status.curFrame;
if SCENE.status.reverse
nextFrame = SCENE.status.curFrame+round(framesToDrop)-1;
else
nextFrame = SCENE.status.curFrame+round(framesToDrop)+1;
end
if nextFrame<=0
if SCENE.status.looped
nextFrame = SCENE.nframes;
else
nextFrame = 1;
end
elseif nextFrame>SCENE.nframes
if SCENE.status.looped
nextFrame = 1;%mod(nextFrame,SCENE.nframes);
else
nextFrame = SCENE.nframes;
end
end
if (framesToDrop<0 && ~SCENE.status.reverse) || (framesToDrop>0 && SCENE.status.reverse)
pause(abs(framesToDrop/SCENE.samplingRate));
end
end
function renewAxisDimensions(bb)
diagonal = sqrt(sum((bb([1,3,5])-bb([2,4,6])).^2))/2;
% center of the boundingbox's bottom
xc = bb(1) + (bb(2) - bb(1))/2;
yc = bb(3) + (bb(4) - bb(3))/2;
% %zc = bb(5) + (bb(6) - bb(5))/2;
axisDimensions = [xc-diagonal xc+diagonal yc-diagonal yc+diagonal];
axis (axisDimensions);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
computeVertices.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/computeVertices.m
| 10,374 |
utf_8
|
62c10be8f0070b96e360ea9b9d7f6212
|
function mot = computeVertices(skel,mot,scale_factor,bonestype)
if isfield(mot,'rotationQuat')
if ~iscell(mot.rotationQuat)
if ~isempty(mot.rotationQuat)
rotQuats = mat2cell(mot.rotationQuat,4*ones(1,numel(mot.animated)));
mot.rotationQuat = cell(mot.njoints,1);
mot.rotationQuat(mot.animated)=rotQuats;
end
end
else
mot.rotationQuat = [];
mot.rotationEuler = [];
end
if (isempty(mot.rotationQuat) || all(cellfun(@(x) isempty(x),mot.rotationQuat)))...
&& (isempty(mot.rotationEuler) || all(cellfun(@(x) isempty(x),mot.rotationEuler)))
if isempty(mot.jointTrajectories)
error('No data available to compute mot.vertices!');
else
mot.faces = [1 2 3;1 3 4;1 4 5;1 5 2;2 3 6;3 4 6;4 5 6;5 2 6];
mot.rotDataAvailable = false;
mot.vertices = cell(mot.njoints,1);
marker_edge_length = 2*scale_factor;
for i=1:mot.njoints
mot.jointTrajectories{i} = mot.jointTrajectories{i} * scale_factor;
mot.vertices{i} = repmat(mot.jointTrajectories{i},6,1);
mot.vertices{i}(2,:) = mot.vertices{i}(2,:)+marker_edge_length;
mot.vertices{i}(6,:) = mot.vertices{i}(6,:)-marker_edge_length;
mot.vertices{i}(7,:) = mot.vertices{i}(7,:)-marker_edge_length;
mot.vertices{i}(12,:) = mot.vertices{i}(12,:)+marker_edge_length;
mot.vertices{i}(13,:) = mot.vertices{i}(13,:)+marker_edge_length;
mot.vertices{i}(17,:) = mot.vertices{i}(17,:)-marker_edge_length;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%hashim
% [mot.jointTrajectories,mot.vertices,mot.faces] = iterativeForwKinematics_local(skel,mot,bonestype);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%hashim
end
else
if isempty(mot.rotationQuat)
mot = convert2quat(skel,mot);
end
mot.rotDataAvailable = true;
mot.rotationQuat(mot.unanimated)={[ones(1,mot.nframes);zeros(3,mot.nframes)]};
[mot.jointTrajectories,mot.vertices,mot.faces] = iterativeForwKinematics_local(skel,mot,bonestype);
% mot.vertices = recursive_forwardKinematicsQuat_local(skel,mot,1,...
% [zeros(15,mot.nframes); mot.rootTranslation],...
% C_quatmult(repmat(skel.rootRotationalOffsetQuat,1,mot.nframes),mot.rotationQuat{1}));
% mot.jointTrajectories = cellfun(@(x) x(end-2:end,:), mot.vertices,'UniformOutput',0);
end
end
function [jointTrajectories,vertices,faces] = H_iterativeForwKinematics_local(skel,mot,bonestype)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%hashim
localSystems = cell(skel.njoints,1);
jointTrajectories = cell(skel.njoints,1);
jointTrajectories{1} = mot.rootTranslation;
vertices = cell(skel.njoints,1);
for i=1:size(skel.paths,1)
for j=2:numel(skel.paths{i})
joint = skel.paths{i}(j);
pred = skel.paths{i}(j-1);
localSystems{joint,1} = localSystems{pred};
jointTrajectories{joint} = mot.jointTrajectories{pred};
[v,faces,nrOfV] = computeVertices_local(skel.nodes(joint).offset,bonestype);
v = reshape(v,3,nrOfV);
vertices{joint} = zeros(nrOfV*3,mot.nframes);
for k=1:nrOfV
vertices{joint}(k*3-2:k*3,:) = mot.jointTrajectories{pred};
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%hashim
end
function [jointTrajectories,vertices,faces] = iterativeForwKinematics_local(skel,mot,bonestype)
localSystems = cell(skel.njoints,1);
localSystems{1} = mot.rotationQuat{1};
jointTrajectories = cell(skel.njoints,1);
jointTrajectories{1} = mot.rootTranslation;
vertices = cell(skel.njoints,1);
for i=1:size(skel.paths,1)
for j=2:numel(skel.paths{i})
joint = skel.paths{i}(j);
pred = skel.paths{i}(j-1);
if isempty(mot.rotationQuat{joint})
localSystems{joint,1} = localSystems{pred};
else
localSystems{joint,1} = C_quatmult(double(real(localSystems{pred})),double(real(mot.rotationQuat{joint})));
end
jointTrajectories{joint} = jointTrajectories{pred}...
+ C_quatrot(skel.nodes(joint).offset,localSystems{joint});
[v,faces,nrOfV] = computeVertices_local(skel.nodes(joint).offset,bonestype);
v = reshape(v,3,nrOfV);
vertices{joint} = zeros(nrOfV*3,mot.nframes);
for k=1:nrOfV
vertices{joint}(k*3-2:k*3,:) = jointTrajectories{pred}...
+ C_quatrot(v(:,k),localSystems{joint});
end
end
end
end
% % function trajectories = recursive_forwardKinematicsQuat_local(skel, mot, node_id, current_position, current_rotation,trajectories)
% %
% % trajectories{node_id,1} = current_position;
% % vertices_mot = zeros(18,mot.nframes);
% %
% % for child_id = skel.nodes(node_id).children'
% %
% % child = skel.nodes(child_id);
% % if (~isempty(mot.rotationQuat{child_id}))
% % child_rotation = quatmult(current_rotation,mot.rotationQuat{child_id});
% % else
% % child_rotation = current_rotation;
% % end
% %
% % % child_rotation = quatmult(current_rotation,mot.rotationQuat{child_id});
% % mot.vertices = computeVertices_local(child.offset);
% %
% % c=1;
% % for i=1:size(mot.vertices,2)
% % vertices_mot(c:c+2,:) = C_quatrot(mot.vertices(:,i),child_rotation);
% % c=c+3;
% % end
% %
% % child_position = vertices_mot + repmat(current_position(16:18,:),6,1);
% % trajectories = recursive_forwardKinematicsQuat_local(skel, mot, child_id, child_position, child_rotation, trajectories);
% % end
% %
% % end
function [vertices,faces,nrOfVertices] = computeVertices_local(child_offset,bonestype)
switch bonestype
case 'diamonds'
child_length = sqrt(sum(child_offset.^2));
dir1 = cross(child_offset,[0;1;0]);
dir1 = dir1/sqrt(sum(dir1.^2));
dir2 = cross(child_offset,dir1);
dir2 = dir2/sqrt(sum(dir2.^2));
off1 = dir1*child_length/10;
off2 = dir2*child_length/10;
centerOfBone = child_offset/4;
vertices = [[0;0;0],...
centerOfBone+off1,...
centerOfBone+off2,...
centerOfBone-off1,...
centerOfBone-off2,...
child_offset];
faces = [1 2 3;1 3 4;1 4 5;1 5 2;2 3 6;3 4 6;4 5 6;5 2 6];
nrOfVertices = 6;
case 'sticks'
sidelength = 3;
dir1 = cross(child_offset,[0;1;0])/2*sidelength;
if ~any(dir1)
dir1=[1;0;0];
else
dir1 = dir1/sqrt(sum(dir1.^2))/2*sidelength;
end
dir2 = cross(child_offset,dir1)/2*sidelength;
if ~any(dir2)
dir2=[1;0;0];
else
dir2 = dir2/sqrt(sum(dir2.^2))/2*sidelength;
end
vertices = [dir1+dir2;...
-dir1+dir2;...
-dir1-dir2;...
dir1-dir2;...
dir1+dir2+child_offset;...
-dir1+dir2+child_offset;...
-dir1-dir2+child_offset;...
dir1-dir2+child_offset];
nrOfVertices = 8;
faces = [1 2 3 4;1 2 6 5;2 3 7 6;3 4 8 7;1 4 8 5; 5 6 7 8];
case 'tubes'
sidelength = 3;
spheresize = 5;
child_offset_n=child_offset/sqrt(sum(child_offset.^2));
rotaxis = cross([0 0 1],child_offset_n);
rotangle = acos(dot(child_offset_n,[0 0 1]));
if child_offset_n(3)==1%~any(rotaxis)
rotaxis=[0;0;1];
rotangle = 0;
end
nn=16;
l = norm(child_offset);
[s.x,s.y,s.z] = cylinder(sidelength,nn);
s.z(2,:)=s.z(2,:)*l;
s.x=s.x(:);
s.y=s.y(:);
s.z=s.z(:);
vertices = zeros((nn+1)*2*3,1);
vertices(1:3:end) = s.x;
vertices(2:3:end) = s.y;
vertices(3:3:end) = s.z;
sina = sin(rotangle/2);
q = [cos(rotangle/2);rotaxis(1)*sina;rotaxis(2)*sina;rotaxis(3)*sina];
vertices = reshape(vertices,3,(nn+1)*2);
vertices = quatrot(vertices,q);
vertices = reshape(vertices,1,(nn+1)*2*3);
faces = zeros(nn,4);
for f=1:nn
ul = f*2-1;
faces(f,:)=[ul ul+2 ul+3 ul+1];
end
[s.x,s.y,s.z] = sphere(nn);
svertices = zeros((nn+1)*(nn+1)*3,1);
svertices(1:3:end) = s.x(:)*spheresize;
svertices(2:3:end) = s.y(:)*spheresize;
svertices(3:3:end) = s.z(:)*spheresize;
sfaces = zeros((nn)*(nn),4);
for c = 1:nn
for r = 1:nn
ul = r+(nn+1)*(c-1);
sfaces((c-1)*nn+r,:)=[ul ul+nn+1 ul+nn+2 ul+1];
end
end
faces = [faces;sfaces+numel(vertices)/3];
vertices = [vertices svertices'];
nrOfVertices = numel(vertices)/3;
case 'spheres'
nn = 16;
spheresize = 3;
[s.x,s.y,s.z] = sphere(nn);
svertices = zeros((nn+1)*(nn+1)*3,1);
svertices(1:3:end) = s.x(:)*spheresize;
svertices(2:3:end) = s.y(:)*spheresize;
svertices(3:3:end) = s.z(:)*spheresize;
sfaces = zeros((nn)*(nn),4);
for c = 1:nn
for r = 1:nn
ul = r+(nn+1)*(c-1);
sfaces((c-1)*nn+r,:)=[ul ul+nn+1 ul+nn+2 ul+1];
end
end
faces = sfaces;
vertices = svertices';
nrOfVertices = numel(vertices)/3;
otherwise
error('Unknown type of bones!');
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
MPP_start.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayerPro/MPP_start.m
| 20,566 |
utf_8
|
a09f5cc60a6440b124c3867424b8d4c7
|
function MPP_start()
global SCENE;
% help --------------------------------------------------------------------
helpDlg = {'Key-Bindings:',...
'------------------------------------------------------',...
'<space>:',' play/pause',...
'<leftarrow>:',' move motion back 10 frames',...
'<downarrow>:',' move motion back 100 frames',...
'<rightarrow>:',' move motion forward 10 frames',...
'<uparrow>:',' move motion forward 100 frames',...
'<shift>+<mouse>:',' orbit rotate scene',...
'<alt>+<mouse>:',' pan scene',...
'<mousewheel>:',' zoom scene'};
% figure and camera settings ----------------------------------------------
axisDimensions = renewAxisDimensions(SCENE.boundingBox);
axis(axisDimensions);
%% control panel-----------------------------------------------------------
SCENE.handles.control_Panel = uipanel(...
'Parent',SCENE.handles.fig,...
'Units','pixels',...
'Position',[2 2 799 110],...
'BackgroundColor',[.97 .97 .97]);
SCENE.handles.goto_First_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.goto_First,...'String','|<',...
'Units','pixels',...
'Position',[2 80 27 20],...
'TooltipString','go to first frame',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@gotoFirstFunction);
SCENE.handles.play_reverse_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.play_reverse,...'String','<|',...
'Units','pixels',...
'Position',[30 80 27 20],...
'TooltipString','play backwards',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@playReverseFunction);
SCENE.handles.pause_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.pause,...'String','||',...
'Units','pixels',...
'Position',[58 80 27 20],...
'TooltipString','pause',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@pauseFunction);
SCENE.handles.play_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.play,...'String','|>',...
'Units','pixels',...
'Position',[86 80 27 20],...
'TooltipString','play',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@playFunction);
SCENE.handles.goto_Last_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.goto_Last,...'String','>|',...
'Units','pixels',...
'Position',[114 80 27 20],...
'TooltipString','go to last frame',...
'BackgroundColor',SCENE.colors.buttons_group1, ...
'CallBack',@gotoLastFunction);
SCENE.handles.slower_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.slower,...'String','<<',...
'Units','pixels',...
'Position',[148 80 27 20],...
'TooltipString','slower',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@slowerFunction);
SCENE.handles.loop_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.unlooped,...'String','|--|',...
'Units','pixels',...
'Position',[176 80 27 20],...
'TooltipString','loop',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@loopFunction);
SCENE.handles.faster_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.faster,...'String','>>',...
'Units','pixels',...
'Position',[204 80 27 20],...
'TooltipString','faster',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@fasterFunction);
SCENE.handles.drawCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.coords+0.5,...
'Units','pixels',...
'Position',[236 80 27 20],...
'TooltipString','draw coordinate system',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawCoordinateSystem);
SCENE.handles.drawLocalCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.localcoords+0.5,...
'Units','pixels',...
'Position',[264 80 27 20],...
'TooltipString','draw local coordinate systems',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawLocalCoordinateSystems2);
SCENE.handles.drawGroundPlane_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.groundPlane,...
'Units','pixels',...
'Position',[292 80 27 20],...
'TooltipString','hide ground plane',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawGroundPlane);
hold all;
computeGroundPlane(SCENE.boundingBox);
if ~SCENE.status.groundPlane_drawn
set(SCENE.handles.drawGroundPlane_Button,'CData',SCENE.buttons.groundPlane+0.5,'TooltipString','draw ground plane');
set(SCENE.handles.groundPlane,'FaceAlpha',0,'EdgeColor','none');
end
hold off;
SCENE.handles.drawJointIDs_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.jointIDs+0.5,...
'Units','pixels',...
'Position',[320 80 27 20],...
'TooltipString','show joint IDs',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawJointIDs);
SCENE.handles.drawSensorCoordSyst_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.sensorcoords+0.5,...
'Units','pixels',...
'Position',[348 80 27 20],...
'TooltipString','draw sensor coordinate systems',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@drawSensorCoordinateSystems);
% SCENE.handles.drawSensorCoordSyst2_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'CData',SCENE.buttons.sensorcoords+0.5,...
% 'Units','pixels',...
% 'Position',[348 80 27 20],...
% 'TooltipString','draw sensor coordinate systems',...
% 'BackgroundColor',SCENE.colors.buttons_group2, ...
% 'CallBack',@drawSensorCoordinateSystems2);
SCENE.handles.spread_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.spread+0.5,...
'Units','pixels',...
'Position',[376 80 27 20],...
'TooltipString','spread motions',...
'BackgroundColor',SCENE.colors.buttons_group2, ...
'CallBack',@spreadFunction);
if SCENE.nmots==1
set(SCENE.handles.spread_Button,'Visible','off');
end
SCENE.handles.MotName_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'String','Name',...
'Units','pixels',...
'Position',[SCENE.size(1)-41-7*32 7 30 20],...
'TooltipString','Display Mot Names',...
'BackgroundColor',[0.0 1.0 0.0], ...
'CallBack',@dispMotNames);
SCENE.handles.Sketch_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'String','Sketch',...
'Units','pixels',...
'Position',[SCENE.size(1)-41-6*32 7 30 20],...
'TooltipString','sketch a frame',...
'BackgroundColor',[0.0 0.1 0.9], ...
'CallBack',@sketchFrame);
SCENE.handles.AutoCam_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.renderScene,...
'Units','pixels',...
'Position',[SCENE.size(1)-41-5*32 7 30 20],...
'TooltipString','compute Auto Camera',...
'BackgroundColor',[0.1 0.8 0.1], ...
'CallBack',@computeAutoCam);
SCENE.handles.AutoCam_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.exportFrame,...
'Units','pixels',...
'Position',[SCENE.size(1)-41-4*32 7 30 20],...
'TooltipString','export motions to obj files',...
'BackgroundColor',[0.1 0.8 0.1], ...
'CallBack',@exportMotToObj);
SCENE.handles.export_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.exportFrame,...
'Units','pixels',...
'Position',[SCENE.size(1)-41-3*32 7 30 20],...
'TooltipString','export frame to obj files',...
'BackgroundColor',[0.9 0.3 0], ...
'CallBack',@exportObjFiles);
SCENE.handles.render_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.renderScene,...
'Units','pixels',...
'Position',[SCENE.size(1)-41-2*32 7 30 20],...
'TooltipString','render Scene to avi',...
'BackgroundColor',[0.9 0.3 0], ...
'CallBack',@renderMPProScene);
SCENE.handles.help_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'String','Help',...
'Units','pixels',...
'FontWeight','bold',...
'HorizontalAlignment','center',...
'Position',[SCENE.size(1)-41-32 6 30 22],...
'CallBack',@helpButtonFunction);
SCENE.handles.quit_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
'CData',SCENE.buttons.quit,...'String','Quit',...
'Units','pixels',...
'Position',[SCENE.size(1)-41 7 30 20],...
'TooltipString','quit',...
'BackgroundColor',[0.9,.0,.0], ...
'CallBack',@closeFunction);
% SCENE.handles.axis_x_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','x',...
% 'Units','pixels',...
% 'Position',[284 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to x',...
% 'BackgroundColor',[0.8,0.8,0.8], ...
% 'CallBack',@setMainAxisFunction);
%
% SCENE.handles.axis_y_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','y',...
% 'Units','pixels',...
% 'Position',[306 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to y',...
% 'BackgroundColor',[0.9,0.9,0.97], ...
% 'CallBack',@setMainAxisFunction);
%
% SCENE.handles.axis_z_Button = uicontrol(SCENE.handles.control_Panel,'Style','Pushbutton', ...
% 'String','z',...
% 'Units','pixels',...
% 'Position',[328 80 20 20],...
% 'FontWeight','bold',...
% 'TooltipString','set main axis to z',...
% 'BackgroundColor',[0.8,0.8,0.8], ...
% 'CallBack',@setMainAxisFunction);
SCENE.handles.status_Panel = uipanel(...
'Parent',SCENE.handles.control_Panel,'Units','pixels',...
'Position',[2 2 SCENE.size(1)-8 30],...
'BackgroundColor',[.97 .97 .97]);
SCENE.handles.curFrameLabel = uicontrol(SCENE.handles.status_Panel,'Style','Text', ...
'String',sprintf(' 1 / %d (%.2f s)', SCENE.nframes,0),...
'Units','pixels',...
'TooltipString','current frame',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[1 0 130 22]);
SCENE.handles.curSpeedLabel = uicontrol(SCENE.handles.status_Panel,'Style','Text', ...
'String','x 1.000',...
'Units','pixels',...
'TooltipString','current speed',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[150 0 40 22]);
% add frame markers above slider ------------------------------------------
if(SCENE.nframes > 1)
SCENE.handles.sliderHandle = uicontrol(SCENE.handles.control_Panel,'Style','Slider', ...
'String','Current Frame',...
'Units','pixels',...
'Max',SCENE.nframes,...
'Min',1,...
'Value',1,...
'SliderStep',[1/SCENE.nframes (1/SCENE.size(1))*40],...
'Position',[2 35 SCENE.size(1)-8 20],...
'BackgroundColor',[.8 .8 .8], ...
'CallBack',@moveFrameSliderFunction);
if SCENE.nframes<=20
numMarks = SCENE.nframes;
else
numMarks = 15;
end
for i=20:-1:5
if mod(SCENE.nframes-1,i)==0
numMarks=i+1;
break;
end
end
posFromLeft = 11;
posFromRight = SCENE.size(1)-62;
posFromLeft = posFromLeft-(posFromRight-posFromLeft)/(SCENE.nframes-1);
for frameNum = 1:(SCENE.nframes-1)/(numMarks-1):SCENE.nframes
uicontrol(SCENE.handles.control_Panel,'Style','Text',...
'String',round(frameNum),'Units','pixels',...
'FontSize',7,'BackgroundColor',[.97 .97 .97],...
'Position',[posFromLeft+(round(frameNum)/SCENE.nframes)*(posFromRight-posFromLeft) 60 45 12]);
end
end
%% callback functions -----------------------------------------------------
function drawGroundPlane(varargin)
if SCENE.status.groundPlane_drawn
SCENE.status.groundPlane_drawn = false;
% set(SCENE.handles.groundPlane,'Visible','off');
set(SCENE.handles.drawGroundPlane_Button,'CData',SCENE.buttons.groundPlane+0.5,'TooltipString','draw ground plane');
set(SCENE.handles.groundPlane,'FaceAlpha',0,'EdgeColor','none');
else
SCENE.status.groundPlane_drawn = true;
% set(SCENE.handles.groundPlane,'Visible','on');
set(SCENE.handles.drawGroundPlane_Button,'CData',SCENE.buttons.groundPlane,'TooltipString','hide ground plane');
set(SCENE.handles.groundPlane,'FaceAlpha',0.7,'EdgeColor','none');
end
end
function playReverseFunction(varargin)
if (SCENE.status.curFrame == 1)
SCENE.status.curFrame = SCENE.nframes;
end
SCENE.status.reverse = true;
SCENE.status.running = true;
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
while SCENE.status.running && (SCENE.status.curFrame>1 || SCENE.status.looped)
nextFrame = getNextFrame_local();
if SCENE.status.running
setFramePro(nextFrame);
drawnow;
SCENE.status.timeStamp = SCENE.timeOffset-toc*SCENE.status.speed;
end
end
SCENE.status.running = false;
end
function pauseFunction(varargin)
SCENE.status.running = false;
end
function playFunction(varargin)
if (SCENE.status.curFrame == SCENE.nframes)
SCENE.status.curFrame = 1;
end
SCENE.status.reverse = false;
SCENE.status.running = true;
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
while SCENE.status.running && (SCENE.status.curFrame<SCENE.nframes || SCENE.status.looped)
nextFrame = getNextFrame_local();
if SCENE.status.running
setFramePro(nextFrame);
drawnow;
SCENE.status.timeStamp = SCENE.timeOffset+toc*SCENE.status.speed;
end
end
SCENE.status.running = false;
end
function gotoFirstFunction(varargin)
SCENE.status.running = false;
SCENE.status.curFrame = 1;
setFramePro(1);
drawnow();
end
function gotoLastFunction(varargin)
SCENE.status.running = false;
SCENE.status.curFrame = SCENE.nframes;
setFramePro(SCENE.nframes);
drawnow();
end
function closeFunction(varargin)
SCENE.status.running = false;
close;
SCENE.objects = [];
SCENE.mots = [];
SCENE.nmots = 0;
SCENE.skels = [];
SCENE.nskels = 0;
% clear global SCENE;
end
function slowerFunction(varargin)
if (SCENE.status.speed > 0.125)
SCENE.status.speed = SCENE.status.speed/2;
set(SCENE.handles.curSpeedLabel,'String',...
sprintf('x %1.3f',SCENE.status.speed));
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
end
end
function fasterFunction(varargin)
if (SCENE.status.speed < 8)
SCENE.status.speed = SCENE.status.speed*2;
set(SCENE.handles.curSpeedLabel,'String',...
sprintf('x %1.3f',SCENE.status.speed));
SCENE.status.timeStamp = (SCENE.status.curFrame)/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
tic;
end
end
function loopFunction(varargin)
if (SCENE.status.looped)
SCENE.status.looped = false;
set(SCENE.handles.loop_Button, 'CData',SCENE.buttons.unlooped,'TooltipString','loop');
else
SCENE.status.looped = true;
set(SCENE.handles.loop_Button, 'CData',SCENE.buttons.looped,'TooltipString','no loop');
end
end
function drawJointIDs(varargin)
if SCENE.status.jointIDs_drawn
SCENE.status.jointIDs_drawn = false;
for ii=1:SCENE.nmots
arrayfun(@(x) delete(x), SCENE.mots{ii}.jointID_handles);
end
set(SCENE.handles.drawJointIDs_Button,'CData',SCENE.buttons.jointIDs+0.5,'TooltipString','show joint IDs');
else
SCENE.status.jointIDs_drawn = true;
if ~SCENE.status.running
setFramePro(SCENE.status.curFrame);
end
set(SCENE.handles.drawJointIDs_Button,'CData',SCENE.buttons.jointIDs,'TooltipString','hide joint IDs');
end
end
function moveFrameSliderFunction(varargin)
if (SCENE.status.running)
SCENE.status.running = false;
end
curFrame = round(get(SCENE.handles.sliderHandle,'Value'));
SCENE.status.timeStamp = curFrame/SCENE.samplingRate;
SCENE.timeOffset = SCENE.status.timeStamp;
setFramePro(curFrame);
drawnow;
end
function spreadFunction(varargin)
if SCENE.status.spread == false
for m=1:SCENE.nmots
if SCENE.mots{m}.rotDataAvailable
spreadVertices(m);
else
spreadVerticesPC(m);
end
end
computeBoundingBoxSCENE();
set(SCENE.handles.spread_Button, 'CData',SCENE.buttons.spread,'TooltipString','unspread motions');
SCENE.status.spread = true;
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
axisDimensions = renewAxisDimensions(SCENE.boundingBox);
axis(axisDimensions);
else
for m=1:SCENE.nmots
if SCENE.mots{m}.rotDataAvailable
unspreadVertices(m);
end
end
computeBoundingBoxSCENE();
set(SCENE.handles.spread_Button, 'CData',SCENE.buttons.spread+0.5,'TooltipString','spread motions');
SCENE.status.spread = false;
if SCENE.status.groundPlane_drawn
computeGroundPlane(SCENE.boundingBox);
end
axisDimensions = renewAxisDimensions(SCENE.boundingBox);
axis(axisDimensions);
end
setFramePro(SCENE.status.curFrame);
end
function helpButtonFunction(src,evnt)
% msgbox(helpDlg,'Help','help');
helpFig = figure( 'Visible','on',...
'Name','Help',...
'NumberTitle','off',...
'Menu','none',...
'Position',[400,200,250,400],...
'Resize', 'on', ...
'Color',[.92 .95 .95]);
uicontrol(helpFig,'Style','Text', ...
'String',helpDlg,...
'Units','pixels',...
'HorizontalAlignment','left',...
'BackgroundColor',[.97 .97 .97],...
'Position',[0 0 250 400]);
end
end
%% local functions --------------------------------------------------------
function nextFrame = getNextFrame_local()
global SCENE;
framesToDrop = SCENE.status.timeStamp*SCENE.samplingRate-SCENE.status.curFrame;
if SCENE.status.reverse
nextFrame = SCENE.status.curFrame+round(framesToDrop)-1;
else
nextFrame = SCENE.status.curFrame+round(framesToDrop)+1;
end
if nextFrame<=0
if SCENE.status.looped
nextFrame = SCENE.nframes;
else
nextFrame = 1;
end
elseif nextFrame>SCENE.nframes
if SCENE.status.looped
nextFrame = 1;%mod(nextFrame,SCENE.nframes);
else
nextFrame = SCENE.nframes;
end
end
if (framesToDrop<0 && ~SCENE.status.reverse) || (framesToDrop>0 && SCENE.status.reverse)
pause(abs(framesToDrop/SCENE.samplingRate));
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildDB.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/parser/buildDB.m
| 19,707 |
utf_8
|
f19c88acf9d2fa4fee5ab29432f55770
|
function res = buildDB(varargin)
% optional settings -------------------------------------------------------
origFrameRate = 119.88;
minFrameNumber = 30;%origFrameRate; % ensure that no motions shorter than minFrameNumber frames are chosen
% HINWEIS: ohne "Orig" sind das die Daten nach "fitRootOrientationsFrameWise"!
options.addPosOrig = 1;
options.addPos = 1;
options.addVelOrig = 1;
options.addVel = 1;
options.addVelL = 1;
options.addAccOrig = 1;
options.addAcc = 1;
options.addAccL = 1;
options.addQuat = 1;
options.addEuler = 0;
options.addInvRootRot = 1;
options.addDeltaRootPos = 1;
options.addDeltaRootOri = 1;
options.addOrigRootPos = 1; % needed for extractMotFromDBmat
options.addOrigRootOri = 1; % needed for extractMotFromDBmat
options.addSkel = 1;
options.addSensorAccs = 1; % calibrated based on t-pose assumed in first frame of motion
options.addMirrors = 0;
options.correctWristOrientations = 1; % only affects lwrist and rwrist quats and thus velL and accL
options.addFootprints = 0;
res_args = checkargins_local(varargin,origFrameRate);
global VARS_GLOBAL;
% Precomputing number of frames and files ---------------------------------
FilesOrFolders = uipickfiles();
pause(0.5);
fprintf('Precomputing total number of frames: \t\t\t\t\t\t');
[totalNrOfFrames,totalNrOfAMCFiles,totalNrOfMATFiles] = ...
countFiles_local(FilesOrFolders,minFrameNumber,origFrameRate,res_args.newFrameRate);
fprintf(' ...finished.');
totalNrOfFiles = totalNrOfAMCFiles+totalNrOfMATFiles;
if totalNrOfMATFiles>0
fprintf('\nTotal number of frames could not be computed because of missing amc-files.\n');
fprintf('However, %i mat-files were found.\n',totalNrOfFiles);
fprintf('Please enter number of FILES (Press Enter for %i): ',totalNrOfFiles);
tmp = input('');
if ~isempty(tmp),
totalNrOfFiles = tmp;
else
fprintf('\b%i\n',totalNrOfFiles);
end
fprintf('Please enter number of FRAMES (%i were counted in %i files): ',totalNrOfFrames,totalNrOfAMCFiles);
totalNrOfFrames = input('');
if isempty(totalNrOfFrames);
totalNrOfFrames=1;
end
end
if options.addMirrors
totalNrOfFrames = 2*totalNrOfFrames;
totalNrOfFiles = 2*totalNrOfFiles;
end
% -------------------------------------------------------------------------
if options.addPosOrig, res.posOrig = single(zeros(93,totalNrOfFrames)); end
if options.addPos, res.pos = single(zeros(93,totalNrOfFrames)); end
if options.addVelOrig, res.velOrig = single(zeros(93,totalNrOfFrames)); end
if options.addVel, res.vel = single(zeros(93,totalNrOfFrames)); end
if options.addVelL, res.velL = single(zeros(93,totalNrOfFrames)); end
if options.addAccOrig, res.accOrig = single(zeros(93,totalNrOfFrames)); end
if options.addAcc, res.acc = single(zeros(93,totalNrOfFrames)); end
if options.addAccL, res.accL = single(zeros(93,totalNrOfFrames)); end
if options.addQuat, res.quat = zeros(116,totalNrOfFrames); end
if options.addEuler, res.euler = single(zeros(59,totalNrOfFrames)); end
if options.addInvRootRot, res.invRootRot = zeros(4,totalNrOfFrames); end
if options.addDeltaRootPos, res.deltaRootPos = zeros(3,totalNrOfFrames); end
if options.addDeltaRootOri, res.deltaRootOri = zeros(4,totalNrOfFrames); end
if options.addOrigRootPos, res.origRootPos = zeros(3,totalNrOfFrames); end
if options.addOrigRootOri, res.origRootOri = zeros(4,totalNrOfFrames); end
if options.addSkel, res.skels = cell(totalNrOfFiles,1); end
if options.addSensorAccs,
sensors = defaultSensors();
sensors = fieldnames(sensors);
nrOfSensors = numel(sensors);
res.sensorAccs = cell2struct(mat2cell(zeros(nrOfSensors*3,totalNrOfFrames),ones(1,nrOfSensors)*3,totalNrOfFrames),sensors);
end
if options.addFootprints
res.footprints = false(2,totalNrOfFrames);
res.dbAnnotation = false(1,totalNrOfFiles);
% Java class for communication with db server
annoqpath = fullfile('..','projects','dynamicMotionGraph','footprints','AnnotationQuery.jar');
postgresspath = fullfile('..','projects','dynamicMotionGraph','footprints','postgresql-8.4-701.jdbc4.jar');
javaaddpath (annoqpath);
javaaddpath (postgresspath);
import AnnotationQuery.*
aq = AnnotationQuery();
end
res.motNames = cell(totalNrOfFiles,1);
res.motStartIDs = single(zeros(totalNrOfFiles,1));
% -------------------------------------------------------------------------
res.frameRate = res_args.newFrameRate;
c = 0;
frame = 1;
for i=1:length(FilesOrFolders)
if isdir(FilesOrFolders{i})
files = dir(fullfile(FilesOrFolders{i},'*.amc'));
if isempty(files), files = dir(fullfile(FilesOrFolders{i},'*.mat')); end
asffiles = dir(fullfile(FilesOrFolders{i},'*.asf'));
pathstr = FilesOrFolders{i};
nrOfFiles = length(files);
else
[pathstr, name, ext] = fileparts(FilesOrFolders{i});
asffiles = dir(fullfile(pathstr,'*.asf'));
clear files;
files(1).name = [name ext];
nrOfFiles = 1;
end
asffiles = arrayfun(@(x) x.name(1:end-4),asffiles,'UniformOutput',0);
for j=1:nrOfFiles
if strcmp(files(j).name(end-2:end),'mat')
amc_filename = files(j).name(1:end-4);
else
amc_filename = files(j).name;
end
for f=1:size(asffiles,1)
if findstr(asffiles{f},amc_filename)
asf_filename = [asffiles{f} '.asf'];
break
end
end
mat_filename = [amc_filename '.mat'];
h = fopen(fullfile(pathstr,mat_filename));
if (h~=-1)
fclose(h);
load(fullfile(pathstr,mat_filename), 'skel', 'mot');
else
try
fprintf('\n');
skel = readASF(fullfile(pathstr,asf_filename));
mot = readAMC(fullfile(pathstr,amc_filename),skel);
save(fullfile(pathstr,mat_filename), 'skel', 'mot');
fprintf('Frames read: \t\t\t\t\t\t\t\t\t\t\t\t\t');
catch
fprintf('\nCould not read file!\n');
end
end
mot.samplingRate = origFrameRate;
mot.frameTime = 1/origFrameRate;
if mot.nframes>=minFrameNumber
if ~res_args.individualSkels
skel = res_args.FixedSkel;
% mot.jointTrajectories = iterativeForwKinematics(skel,mot);
elseif ~res_args.individualBoneLengths
for k=1:numel(res_args.FixedBoneLengths)
skel.nodes(k).length = res_args.FixedBoneLengths(k);
skel.nodes(k).offset = skel.nodes(k).direction * res_args.FixedBoneLengths(k);
end
% mot.jointTrajectories = iterativeForwKinematics(skel,mot);
end
mot = changeFrameRate(skel,mot,res_args.newFrameRate);
for mm=1:1+options.addMirrors
c = c+1;
if mm==2
[skel,mot] = mirrorMot(skel,mot); % mirrorMot does forward kinematics
res.motNames{c} = [amc_filename '.mirrored'];
else
res.motNames{c} = amc_filename;
end
if options.correctWristOrientations
[mot,res_tmp] = correctOrientationsInMot(skel,mot);
mot.jointTrajectories = iterativeForwKinematics(skel,mot);
end
res.motStartIDs(c) = frame;
if options.addSkel
res.skels{c} = skel;
end
if options.addVelOrig, mot = addVelToMot(mot); end;
if options.addAccOrig, mot = addAccToMot(mot); end;
if options.addFootprints
[mot,res.dbAnnotation(j)] = detectFootprints(skel,mot,aq);
end
sf = frame;
ef = frame+mot.nframes-1;
if options.addPosOrig, res.posOrig(:,sf:ef) = cell2mat(mot.jointTrajectories); end
if options.addVelOrig, res.velOrig(:,sf:ef) = cell2mat(mot.jointVelocities); end
if options.addAccOrig, res.accOrig(:,sf:ef) = cell2mat(mot.jointAccelerations); end
if options.addOrigRootPos, res.origRootPos(:,sf:ef) = mot.rootTranslation; end
if options.addOrigRootOri, res.origRootOri(:,sf:ef) = mot.rotationQuat{1}; end
if mm==1
if options.addFootprints, res.footprints(:,sf:ef) = mot.footprints; end
else
% mirror footprints
if options.addFootprints, res.footprints(:,sf:ef) = flipud(mot.footprints); end
end
if options.addDeltaRootPos
t1 = mot.rootTranslation(:,1:end-1);
t2 = mot.rootTranslation(:,2:end);
res.deltaRootPos(:,sf:ef) = [ [0;0;0] C_quatrot(t2-t1,C_quatinv(mot.rotationQuat{1}(:,2:end)))];
end
if options.addDeltaRootOri
q1 = mot.rotationQuat{1}(:,1:end-1);
q2 = mot.rotationQuat{1}(:,2:end);
res.deltaRootOri(:,sf:ef) = [[1;0;0;0] C_quatmult(C_quatinv(q1),q2)];
end
if options.addPos || options.addVel || options.addAcc || options.addQuat || options.addEuler || options.addAccL || options.addVelL
mot0 = mot;
mot0.rootTranslation(:,:) = 0;
[mot0,qy,~,q_L2G] = fitRootOrientationsFrameWise(skel,mot0);
if options.addInvRootRot
res.invRootRot(:,sf:ef) = qy;
end
if options.addPos
res.pos(:,sf:ef) = cell2mat(mot0.jointTrajectories);
end
if options.addVelL || options.addAccL
q_G2L = cellfun(@(x) C_quatinv(x),q_L2G,'UniformOutput',0);
end
if options.addVel || options.addVelL
mot0 = addVelToMot(mot0);
if options.addVel
res.vel(:,sf:ef) = cell2mat(mot0.jointVelocities);
end
if options.addVelL
res.velL(:,sf:ef) = cell2mat(cellfun(@(x,y) C_quatrot(x,y),mot0.jointVelocities,q_G2L,'UniformOutput',0));
end
end
if options.addAcc || options.addAccL
mot0 = addAccToMot(mot0);
if options.addAcc
res.acc(:,sf:ef) = cell2mat(mot0.jointAccelerations);
end
if options.addAccL
res.accL(:,sf:ef) = cell2mat(cellfun(@(x,y) C_quatrot(x,y),mot0.jointAccelerations,q_G2L,'UniformOutput',0));
end
end
if options.addEuler
mot0 = convert2euler(skel,mot0);
res.euler(:,sf:ef) = real(cell2mat(mot0.rotationEuler(mot0.animated)));
end
if options.addQuat
res.quat(:,sf:ef) = cell2mat(mot0.rotationQuat(mot0.animated));
end
end
if options.addSensorAccs
if ~options.correctWristOrientations || mm==2
res_tmp = simulateLocalAccsFromAMC2(skel,mot);
end
for s=1:nrOfSensors
res.sensorAccs.(sensors{s})(:,sf:ef) = res_tmp.(sensors{s}).acc_L;
end
end
frame = frame + mot.nframes;
fprintf('\b\b\b\b\b\b\b%7i',frame-1);
end
end
end
end
if options.addFootprints
% delete Java object
clear('aq');
end
fprintf(' ...finished.\n');
if frame-1<totalNrOfFrames || c<totalNrOfFiles
res = cleanup_local(res,totalNrOfFiles,c,totalNrOfFrames,frame-1);
res.nrOfFrames = frame-1;
else
res.nrOfFrames = frame-1;
end
end
%% local functions
function db = cleanup_local(db,totalNrOfFiles,actualNrOfFiles,totalNrOfFrames,actualNrOfFrames)
% if c~=totalNrOfFiles
% res.motNames = res.motNames(1:c);
% res.motStartIDs = res.motStartIDs(1:c);
% end
% totalNrOfFrames = db.nrOfFrames;
% totalNrOfFiles = numel(db.motStartIDs);
% mots2keep = 1:actualNrOfFiles;
% frames2keep = 1:actualNrOfFrames;
fields = fieldnames(db);
for f=1:size(fields,1)
field = fields{f};
[a,b]=ismember([totalNrOfFiles,totalNrOfFrames],size(db.(field)));
if all(a)
fprintf('Caution: Field %s is of ambiguous size. Please remove values manually.\n',field);
elseif a(1)
if b(1)==1
% db.(field)=db.(field)(mots2keep,:);
db.(field)(actualNrOfFiles+1:end,:) = [];
elseif b(1)==2
db.(field)(:,actualNrOfFiles+1:end) = [];
% db.(field)=db.(field)(:,mots2keep);
end
elseif a(2)
if b(2)==1
db.(field)(actualNrOfFrames+1:end,:) = [];
% db.(field)=db.(field)(frames2keep,:);
elseif b(2)==2
db.(field)(:,actualNrOfFrames+1:end) = [];
% db.(field)=db.(field)(:,frames2keep);
end
elseif isstruct(db.(field))
db.(field) = cleanup_local(db.(field),totalNrOfFiles,actualNrOfFiles,totalNrOfFrames,actualNrOfFrames);
end
end
end
function isskel = isSkel_local(var)
isskel = isstruct(var) && all(isfield(var,{'njoints','nodes'}));
end
function isframerate = isFrameRate_local(var)
isframerate = isnumeric(var) && numel(var)==1;
end
function isbonelengthsarray = isBoneLengthsArray_local(var)
isbonelengthsarray = isnumeric(var) && numel(var)==31;
end
function [totalNrOfFrames,totalNrOfAMCFiles,totalNrOfMATFiles] = countFiles_local(FilesOrFolders,minFrameNumber,origFrameRate,newFrameRate)
totalNrOfFrames = 0;
totalNrOfAMCFiles = 0;
totalNrOfMATFiles = 0;
for i=1:length(FilesOrFolders)
if isdir(FilesOrFolders{i})
files = dir(fullfile(FilesOrFolders{i},'*.amc'));
nrOfAMCFiles = numel(files);
if nrOfAMCFiles==0
nrOfMATFiles = numel(dir(fullfile(FilesOrFolders{i},'*.mat')));
else
nrOfMATFiles = 0;
end
pathstr = FilesOrFolders{i};
else
[pathstr, name, ext] = fileparts(FilesOrFolders{i});
if strcmp(ext,'.amc')
nrOfAMCFiles = 1;
nrOfMATFiles = 0;
clear files;
files(1).name = [name ext];
elseif strcmp(ext,'.mat'),
if numel((dir(fullfile(pathstr,name))))>0
clear files;
files(1).name = name;
nrOfAMCFiles = 1;
nrOfMATFiles = 0;
else
nrOfMATFiles = 1;
nrOfAMCFiles = 0;
end
end
end
if nrOfAMCFiles>0
for j=1:nrOfAMCFiles
amc_filename = fullfile(pathstr,files(j).name);
nrOfFrames = readNrOfFramesFromFile(amc_filename);
if (nrOfFrames >= minFrameNumber)
nrOfFrames = numel(1:origFrameRate/newFrameRate:nrOfFrames);
totalNrOfFrames = totalNrOfFrames + nrOfFrames;
totalNrOfAMCFiles = totalNrOfAMCFiles + 1;
end
fprintf('\b\b\b\b\b\b\b%7i',totalNrOfFrames);
end
else
totalNrOfMATFiles = totalNrOfMATFiles + nrOfMATFiles;
end
end
end
function res = checkargins_local(argins,origFrameRate)
% Checking nargins --------------------------------------------------------
switch numel(argins)
case 0
res.newFrameRate = origFrameRate;
res.individualSkels = true;
case 1
if isSkel_local(argins{1})
res.individualSkels = false;
res.individualBoneLengths = false;
res.FixedSkel = argins{1};
elseif isFrameRate_local(argins{1})
res.individualSkels = true;
res.individualBoneLengths = true;
res.newFrameRate = argins{1};
elseif isBoneLengthsArray_local(argins{1})
res.individualSkels = true;
res.individualBoneLengths = false;
res.FixedBoneLengths = argins{1};
else
error('Wrong types of argins!');
end
case 2
if isSkel_local(argins{1}) && isFrameRate_local(argins{2})
res.individualSkels = false;
res.individualBoneLengths = false;
res.FixedSkel = argins{1};
res.newFrameRate = argins{2};
elseif isSkel_local(argins{2}) && isFrameRate_local(argins{1})
res.individualSkels = false;
res.individualBoneLengths = false;
res.FixedSkel = argins{2};
res.newFrameRate = argins{1};
elseif isFrameRate_local(argins{1}) && isBoneLengthsArray_local(argins{2})
res.individualSkels = true;
res.individualBoneLengths = false;
res.newFrameRate = argins{1};
res.FixedBoneLengths = argins{2};
elseif isFrameRate_local(argins{2}) && isBoneLengthsArray_local(argins{1})
res.individualSkels = true;
res.individualBoneLengths = false;
res.newFrameRate = argins{2};
res.FixedBoneLengths = argins{1};
else
error('Wrong types of argins!');
end
otherwise
error('Wrong number of argins!');
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
emptyMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/parser/emptyMotion.m
| 6,695 |
utf_8
|
8b76a6adb6c20591b900c33d82bc4bd4
|
function mot = emptyMotion(varargin)
switch nargin
case 0
mot = struct('njoints',0,... % number of joints
'nframes',0,... % number of frames
'frameTime',nan,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',nan,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',cell(1,1),... % 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',cell(1,1),... % cell array of joint names: maps node ID to joint name
'boneNames',cell(1,1),... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',cell(1,1),... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',[],... % vector of IDs for animated joints/bones
'unanimated',[],... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
case 1
if ismot_local(varargin{1})
refmot = varargin{1};
mot = struct('njoints',refmot.njoints,... % number of joints
'nframes',0,... % number of frames
'frameTime',refmot.frameTime,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',refmot.samplingRate,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',[],...cell(refmot.njoints,1),...% 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',[],...{cell(refmot.njoints,1)},...% rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',[],...{cell(refmot.njoints,1)},...% rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',{refmot.jointNames},... % cell array of joint names: maps node ID to joint name
'boneNames',{refmot.boneNames},... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',{refmot.nameMap},... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',refmot.animated,... % vector of IDs for animated joints/bones
'unanimated',refmot.unanimated,... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
elseif isskel_local(varargin{1})
skel = varargin{1};
mot = struct('njoints',skel.njoints,... % number of joints
'nframes',0,... % number of frames
'frameTime',nan,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',nan,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',[],...{cell(skel.njoints,1)},...% 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',[],...{cell(skel.njoints,1)},... % rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',[],...{cell(skel.njoints,1)},... % rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',{skel.jointNames},... % cell array of joint names: maps node ID to joint name
'boneNames',{skel.boneNames},... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',{skel.nameMap},... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',skel.animated,... % vector of IDs for animated joints/bones
'unanimated',skel.unanimated,... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
end
end
end
%% local functions
function ismot_bool = ismot_local(arg)
if isfield(arg,'nframes') && isfield(arg,'rotationQuat')
ismot_bool = true;
else
ismot_bool = false;
end
end
function isskel_bool = isskel_local(arg)
if isfield(arg,'njoints') && isfield(arg,'nodes')
isskel_bool = true;
else
isskel_bool = false;
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
amc_to_matrix.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/parser/ASFAMCparser/amc_to_matrix.m
| 2,671 |
utf_8
|
32da4a50441d0da713b0bb6ea3349c22
|
% Reads data from an AMC motion file into a Matlab matrix variable.
% AMC file has to be in the AMC format used in the online CMU motion capture library.
% number of dimensions = number of columns = 62
% function D = amc_to_matrix(fname)
% fname = name of disk input file, in AMC format
% Example:
% D = amc_to_matrix(fname)
%
% Jernej Barbic
% CMU
% March 2003
% Databases Course
function [D] = amc_to_matrix(fname)
fid=fopen(fname, 'rt');
if fid == -1,
fprintf('Error, can not open file %s.\n', fname);
return;
end;
% read-in header
line=fgetl(fid);
while ~strcmp(line,':DEGREES')
line=fgetl(fid);
end
D=[];
dims =[6 3 3 3 3 3 3 2 3 1 1 2 1 2 2 3 1 1 2 1 2 3 1 2 1 3 1 2 1];
locations = [1 7 10 13 16 19 22 25 27 30 31 32 34 35 37 39 42 43 44 46 47 49 52 53 55 56 59 60 62];
% read-in data
% labels can be in any order
frame=1;
while ~feof(fid)
if rem(frame,100) == 0
disp('Reading frame: ');
disp(frame);
end;
row = zeros(62,1);
% read frame number
line = fscanf(fid,'%s\n',1);
for i=1:29
% read angle label
id = fscanf (fid,'%s',1);
switch (id)
case 'root', index = 1;
case 'lowerback', index = 2;
case 'upperback', index = 3;
case 'thorax', index = 4;
case 'lowerneck', index = 5;
case 'upperneck', index = 6;
case 'head', index = 7;
case 'rclavicle', index = 8;
case 'rhumerus', index = 9;
case 'rradius', index = 10;
case 'rwrist', index = 11;
case 'rhand', index = 12;
case 'rfingers', index = 13;
case 'rthumb', index = 14;
case 'lclavicle', index = 15;
case 'lhumerus', index = 16;
case 'lradius', index = 17;
case 'lwrist', index = 18;
case 'lhand', index = 19;
case 'lfingers', index = 20;
case 'lthumb', index = 21;
case 'rfemur', index = 22;
case 'rtibia', index = 23;
case 'rfoot', index = 24;
case 'rtoes', index = 25;
case 'lfemur', index = 26;
case 'ltibia', index = 27;
case 'lfoot', index = 28;
case 'ltoes', index = 29;
otherwise
fprintf('Error, labels in the amc are not correct.\n');
return;
end
% where to put the data
location = locations(index);
len = dims(index);
if len == 6
x = fscanf (fid,'%f %f %f %f %f %f\n',6);
else
if len == 3
x = fscanf (fid,'%f %f %f\n',3);
else
if len == 2
x = fscanf (fid,'%f %f\n',2);
else
if len == 1
x = fscanf (fid,'%f\n',1);
end
end
end
end
row(location:location+len-1,1) = x;
end
row = row';
D = [D; row];
frame = frame + 1;
end
disp('Total number of frames read: ');
disp(frame-1);
fclose(fid);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
matrix_to_amc.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/parser/ASFAMCparser/matrix_to_amc.m
| 2,225 |
utf_8
|
3d0a06d61f8d69c60c6e78641144ea64
|
% Writes motion data from matrix D to an AMC file on disk.
% The ACM format is the format used in the CMU online motion capture database
% function [] = matrix_to_amc(fname, D)
% fname = output disk file name for AMC file
% D = input Matlab data matrix
% Example:
% matrix_to_amc('running1.amc', D)
%
%
% Jernej Barbic
% CMU
% March 2003
% Databases Course
function [] = matrix_to_amc(fname, D)
fid=fopen(fname, 'wt');
if fid == -1,
fprintf('Error, can not open file %s.\n', fname);
return;
end;
% print header
fprintf(fid,'#!Matlab matrix to amc conversion\n');
fprintf(fid,':FULLY-SPECIFIED\n');
fprintf(fid,':DEGREES\n');
[rows, cols] = size(D);
% print data
for frame=1:rows
fprintf(fid,'%d\n',frame);
fprintf(fid,'root %f %f %f %f %f %f\n', D(frame,1:6));
fprintf(fid,'lowerback %f %f %f\n', D(frame,7:9));
fprintf(fid,'upperback %f %f %f\n', D(frame,10:12));
fprintf(fid,'thorax %f %f %f\n', D(frame,13:15));
fprintf(fid,'lowerneck %f %f %f\n', D(frame,16:18));
fprintf(fid,'upperneck %f %f %f\n', D(frame,19:21));
fprintf(fid,'head %f %f %f\n', D(frame,22:24));
fprintf(fid,'rclavicle %f %f\n', D(frame,25:26));
fprintf(fid,'rhumerus %f %f %f\n', D(frame,27:29));
fprintf(fid,'rradius %f\n', D(frame,30));
fprintf(fid,'rwrist %f\n', D(frame,31));
fprintf(fid,'rhand %f %f\n', D(frame,32:33));
fprintf(fid,'rfingers %f\n', D(frame,34));
fprintf(fid,'rthumb %f %f\n', D(frame,35:36));
fprintf(fid,'lclavicle %f %f\n', D(frame,37:38));
fprintf(fid,'lhumerus %f %f %f\n', D(frame,39:41));
fprintf(fid,'lradius %f\n', D(frame,42));
fprintf(fid,'lwrist %f\n', D(frame,43));
fprintf(fid,'lhand %f %f\n', D(frame,44:45));
fprintf(fid,'lfingers %f\n', D(frame,46));
fprintf(fid,'lthumb %f %f\n', D(frame,47:48));
fprintf(fid,'rfemur %f %f %f\n', D(frame,49:51));
fprintf(fid,'rtibia %f\n', D(frame,52));
fprintf(fid,'rfoot %f %f\n', D(frame,53:54));
fprintf(fid,'rtoes %f\n', D(frame,55));
fprintf(fid,'lfemur %f %f %f\n', D(frame,56:58));
fprintf(fid,'ltibia %f\n', D(frame,59));
fprintf(fid,'lfoot %f %f\n', D(frame,60:61));
fprintf(fid,'ltoes %f\n', D(frame,62));
end
fclose(fid);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
filterR4.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/quaternions/filterR4.m
| 2,217 |
utf_8
|
c795e21bddf40ecb117cc6e70f7f575c
|
function [Y,t] = filterR4(varargin)
% Y = filterR4(w,X,step,padding_method)
% Filters curves embedded in the unit quaternion sphere with a sliding window.
% Simply views quats as 4D data without additional structure and renormalizes
% to S^3 after filtering
%
% Input: w, weight vector
% X, 4xN matrix of input unit quaternions
% optional: step, step size for window (window length is length(w)), default is step=1.
% ->!! Length of output sequence is ceil(N/step). !! <-
% optional: padding method, one of {'symmetric', 'zero'}. default is 'symmetric'
%
% Output: Y, Filtered version of X
% t, running time for filter.
%
switch (nargin)
case 2
w = varargin{1};
X = varargin{2};
step = 1;
padding_method = 'symmetric';
case 3
w = varargin{1};
X = varargin{2};
step = varargin{3};
padding_method = 'symmetric';
case 4
w = varargin{1};
X = varargin{2};
step = varargin{3};
padding_method = varargin{4};
otherwise
error('Wrong number of arguments!');
end
L = size(w,2);
N = size(X,2);
if (L>N)
error('Filter length must not be larger than number of data points!');
end
%%%% prepare data set X by means of pre- and postpadding
switch mod(L,2)
case 0 % even filter length
prepad_length = L/2;
postpad_length = L/2 - 1;
case 1 % odd filter length
prepad_length = (L - 1)/2;
postpad_length = (L - 1)/2;
end
tic;
if (strncmp(padding_method,'symmetric',1))
pre = fliplr(X(:,1:prepad_length));
post = fliplr(X(:,N-postpad_length+1:N));
elseif (strncmp(padding_method,'zero',1))
pre = [ones(1,prepad_length);zeros(3,prepad_length)];
post = [ones(1,postpad_length);zeros(3,postpad_length)];;
else
error('Unknown padding option!');
end
X = [pre X post];
Y = zeros(4,ceil(N/step));
for (i=1:ceil(N/step))
pos = step*(i-1)+1;
Y(:,i) = bruteForceAverageR4(X(:,pos:pos+L-1),w);
end
t = toc;
%%%%%%%%%%%
function Y = bruteForceAverageR4(X,w)
Y = sum(X.*repmat(w,4,1),2);
Y = Y./sqrt(sum(Y.^2,1));
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
isColor.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayer/isColor.m
| 295 |
utf_8
|
e20f1469401335aa9acac9aed8ad5f54
|
%% validator for color
% returns true if x is a 1x3 matrix of doubles between 0.0 and 1.0
function val = isColor(x)
if(...
min(min(x)) >= 0.0 && ...
max(max(x)) <= 1.0 && ...
size(x,1) == 1 && ...
size(x,2) == 3)
val = 1;
else
val = 0;
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
removeDupVerts.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayer/removeDupVerts.m
| 788 |
utf_8
|
8a886ae550cf2eca76d4d94c848de739
|
% remove duplicate vertices and replace indices in patch with new
% vertex indices
function [vertsOut,patchesOut] = removeDupVerts(vertsIn, patchesIn)
patchesOut = patchesIn';
index = [1:size(vertsIn,1);zeros(1,size(vertsIn,1))]';
for i = 1:size(vertsIn,1)
for j = 1:size(vertsIn,1)
if(~index(j,2))
if(vertsIn(i,1) == vertsIn(j,1) &&...
vertsIn(i,2) == vertsIn(j,2) &&...
vertsIn(i,3) == vertsIn(j,3))
index(j,1) = i;
index(j,2) = 1;
end
end
end
end
vertsUsed = intersect(1:size(vertsIn,1), index(:,1));
vertsOut = vertsIn(vertsUsed,:);
for i = 1:size(index,1)
patchesOut(i) = find(vertsUsed == index(i,1));
end
patchesOut = patchesOut';
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
isMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayer/isMot.m
| 1,399 |
utf_8
|
eea71926a8ea8e1bbe6cb280b10db8e5
|
function out = isMot(mot)
% returns true if given argument is a motion-type structure
% this function tests if mot is of type struct and contains the
% following fields
% fieldsRequired = {...
% 'njoints', 'nframes', 'frameTime', 'samplingRate'...
% 'jointTrajectories', 'rootTranslation', 'rotationEuler',...
% 'rotationQuat', 'jointNames', 'boneNames', 'nameMap',...
% 'animated', 'unanimated', 'boundingBox', 'filename',...
% 'documentation', 'angleUnit'};
fieldsRequired = {...
'njoints', 'nframes', 'frameTime', 'samplingRate'...
'jointTrajectories', 'boundingBox', 'filename'};
for i = 1:size(mot,2)
m = mot{1,i};
if(isstruct(m))
fieldsInMot = fieldnames(m);
for f = fieldsRequired
if (any(strcmp(f,fieldsInMot)))
else
out = false;
fprintf('motion is not valid: \n\t field ''%s'' is missing\n', f{1,1});
break;
end
out = true;
end
else
out = false;
fprintf('type mismatch on ''mot'': type struct expected\n');
break;
end
end
if(out == false)
r = '';
for f = fieldsRequired
r = strcat('''',f{1,1},'''',',', r);
end
r = strcat('fields required for mot: \n\t(',r,')\n');
fprintf(r);
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
plotMultiLayerResult2Fig_forVideo.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionplayer/video/plotMultiLayerResult2Fig_forVideo.m
| 7,956 |
utf_8
|
ee68b584f34ba9d1d719815af8bd9f60
|
function plotMultiLayerResult2Fig_forVideo(annotation,hits4File, parameter)
if nargin < 3
parameter = struct;
end
if isfield(parameter, 'printFigure') == 0
parameter.printFigure = 0;
end
if isfield(parameter, 'filenamePrefix') == 0
parameter.filenamePrefix = 'figures/tensorClassification_';
end
if isfield(parameter, 'paperPositionClassification') == 0
parameter.paperPositionClassification = [1,1,6.3,2.5];
end
if isfield(parameter, 'paperPositionPartDTW') == 0
parameter.paperPositionPartDTW = [1,1,6.3,2.5];
end
if isfield(parameter, 'omitTitle') == 0
parameter.omitTitle = 0;
end
if isfield(parameter, 'drawFineAnnotations') == 0
parameter.drawFineAnnotations = 1;
end
if isfield(parameter,'highlightFrame')==0
parameter.highlightFrame = 0;
end
if isfield(parameter,'position')==0
parameter.position = [ 200 200 1000 440];
end
global VARS_GLOBAL;
motClasses = flipud(fieldnames(annotation));
document=hits4File.amc;
docLength = hits4File.orgNumFrames;
figure();
set(gcf, 'renderer', 'painters') ;
set(gcf, 'name', [document]);
[temp, docShortName] = fileparts(document);
set(gcf, 'position', parameter.position)
subplot(7,1,1:4);
set(gcf, 'color', [1, 1, 1]);
%% Collect Annotations for document.
% hitData = classificationResult.(class)((classificationResult.(class)(:,1) == currentDocument), [2:3, 6]);
% h1=subplot(2,1,1);
set(gca, 'ytick', [1.5:2:length(motClasses)*2]);
set(gca, 'yticklabel', motClasses);
% set(gca, 'XTick',1:100:hits4File.orgNumFrames);
% set(gca, 'XTickLabel',0:100:hits4File.orgNumFrames);
axis([0 hits4File.orgNumFrames 0.5 length(motClasses)*2+0.5]);
for mClass=1:size(motClasses,1);
line([1, hits4File.orgNumFrames], [2*mClass+0.5, 2*mClass+0.5]);
hitData=zeros(1,2);
count=1;
if ~isempty(annotation.(motClasses{mClass}))
for i=1:size(annotation.(motClasses{mClass}),1);
if strcmp(strcat(VARS_GLOBAL.dir_root,annotation.(motClasses{mClass})(i,3)),document)
hitData(count,:)=cell2mat(annotation.(motClasses{mClass})(i,1:2));
files{count}=annotation.(motClasses{mClass})(i,3);
count=count+1;
end
end
end
for h=1:size(hitData, 1)
hitStart = hitData(h, 1);
hitEnd = hitData(h, 2);
% remap from 30Hz to 120 Hz
% animationStart = hitStart * 4 - 4 + 1;
% animationEnd = hitEnd * 4 - 4 + 1;
%draw rectangle
color = [1 0 0]; %[0 127 14]/255;
yPosition = mClass*2;
if ~parameter.drawFineAnnotations && parameter.highlightFrame >= hitStart && parameter.highlightFrame <= hitEnd
color = [0,1,1]; % cyan
end
hitWidth = hitStart-hitEnd;
if hitWidth == 0
handle = line([hitStart, hitStart], [yPosition-0.3, yPosition+0.3], 'Color', [0,0,0], 'LineWidth', 1);
else
handle = rectangle('Position', [hitStart, yPosition-0.3,...
hitEnd - hitStart+1, 0.6],...
'EdgeColor', [0 0 0],...
'FaceColor', color,...
'LineWidth', 0.1);
end
% set(handle,'ButtonDownFcn',{@animateRectOnClick,...
% docName,...
% animationStart,...
% animationEnd, ...
% cost,...
% hitStart,...
% hitEnd,...
% });
end
end
if (parameter.drawFineAnnotations)
for hit=1:hits4File.numHits
mClassStr = hits4File.hitProperties{1,hit}.motionClass;
tmp=strfind(motClasses,cell2mat(mClassStr));
mClass=1;
while isempty(cell2mat(tmp(mClass)))
mClass=mClass+1;
end
hitStart=hits4File.startFrames(hit);
hitEnd =hits4File.endFrames (hit);
color = [0 0.7 0];
yPosition = 2*mClass-1;
hitWidth = hitStart-hitEnd;
if hitWidth == 0
handle = line([hitStart, hitStart], [yPosition-0.3, yPosition+0.3], 'Color', [0,0,0], 'LineWidth', 1);
else
if parameter.highlightFrame >= hitStart && parameter.highlightFrame <= hitEnd
color = [0,1,1]; % cyan
end
handle = rectangle('Position', [hitStart, yPosition-0.3,...
hitEnd - hitStart+1, 0.6],...
'EdgeColor', [0 0 0],...
'FaceColor', color,...
'LineWidth', 0.1);
end
end
end
line([docLength, docLength], [0.5, 2*length(motClasses)+0.5], 'Color', [0,0,0]);
line([1, docLength], [2*length(motClasses)+0.5, 2*length(motClasses)+0.5], 'Color', [0,0,0]);
if parameter.highlightFrame > 0
line([parameter.highlightFrame, parameter.highlightFrame], [0.5, 2*length(motClasses)+0.5], 'Color', [0,1,1]);
end
%% Collect PartDTW Hits for Document!
if (parameter.drawFineAnnotations)
subplot(7,1,6:7);
set(gcf, 'renderer', 'painters') ;
set(gcf, 'name', [document]);
%create styleList
[numHits,coeffs,styleList]=countHitsPerFrameAndCoefs(hits4File);
ylabels = renameSubclasses(styleList);
set(gca, 'ytick', 1:1:size(styleList,2));
set(gca, 'yticklabel', ylabels);
% set(gca, 'XTick',1:100:hits4File.orgNumFrames);
% set(gca, 'XTickLabel',0:100:hits4File.orgNumFrames);
for styInd=1:size(styleList,2)
line([1, hits4File.orgNumFrames], [styInd+0.5, styInd+0.5]);
end
axis([0 hits4File.orgNumFrames 0.5 size(styleList,2)+0.5]);
for hit=1:hits4File.numHits
[maxval,maxPos]=max(hits4File.hitProperties{1,hit}.res.coeffsX{1});
subClass=hits4File.hitProperties{1,hit}.styles{maxPos};
tmp=strfind(styleList,subClass);
mClass=1;
while isempty(cell2mat(tmp(mClass)))
mClass=mClass+1;
end
hitStart=hits4File.startFrames(hit);
hitEnd =hits4File.endFrames (hit);
if maxval>0.0
color=[0 0.7 0];
else
color=[0 1 0];
end
yPosition = mClass;
hitWidth = hitStart-hitEnd;
if hitWidth == 0
handle = line([hitStart, hitStart], [yPosition-0.3, yPosition+0.3], 'Color', [0,0,0], 'LineWidth', 1);
else
if parameter.highlightFrame >= hitStart && parameter.highlightFrame <= hitEnd
color = [0,1,1]; % cyan
end
handle = rectangle('Position', [hitStart, yPosition-0.3,...
hitEnd - hitStart+1, 0.6],...
'EdgeColor', [0 0 0],...
'FaceColor', color,...
'LineWidth', 0.1);
set(handle,'ButtonDownFcn',{@motionPlayerCallback2, ...
hits4File.hitProperties{1,hit}.res.skel, ...
hits4File.hitProperties{1,hit}.orgMot, ...
hits4File.hitProperties{1,hit}.res.skel, ...
hits4File.hitProperties{1,hit}.recMotUnWarp});
end
end
line([docLength, docLength], [0.5, 2*length(styleList)+0.5], 'Color', [0,0,0]);
line([1, docLength], [length(styleList)+0.5, length(styleList)+0.5], 'Color', [0,0,0]);
if parameter.highlightFrame > 0
line([parameter.highlightFrame, parameter.highlightFrame], [0.5, 2*length(styleList)+0.5], 'Color', [0,1,1]);
end
end
end
function motionPlayerCallback1(src, eventdata, skel1, mot1)
motionplayer('skel',{skel1}, 'mot', {mot1});
end
function motionPlayerCallback2(src, eventdata, skel1, mot1, skel2, mot2)
motionplayer('skel',{skel1 skel2}, 'mot', {mot1 mot2});
end
function motionPlayerCallbackLoad(src, eventdata, amc, startF, endF)
infos=filename2info(amc);
[skel,mot] =readMocap(fullfile(infos.amcpath,infos.asfname),amc);
mot=cutMotion(mot,startF,endF);
motionplayer('skel',{skel}, 'mot', {mot});
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reverseMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/reverseMotion.m
| 379 |
utf_8
|
645f63d913001e86ef960cb567c449c1
|
% function reverseMotion
% author: Jochen Tautges ([email protected])
function mot = reverseMotion(skel,mot)
mot.rotationQuat = cellfun(@(x) fliplr(x),mot.rotationQuat,'UniformOutput',0);
mot.filename = [mot.filename '.reversed'];
mot.rootTranslation = fliplr(mot.rootTranslation);
mot.jointTrajectories = iterativeForwKinematics(skel,mot);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
fitMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/fitMotion.m
| 1,137 |
utf_8
|
6699036712fc3b7a111e64eebc7a4d0b
|
% FUNCTION fitMotion applays a translation and a rotation around y-axis to
% a given motion. The motion is moved with root position to the origin with
% the first frame. And rotated that the main direction goes along the
% x-Axis.
% INPUT:
% skel: struct: skeleton definition for the motion
% mot: struct:
function [mot,angle,x0,z0]=fitMotion(skel,mot)
% Spatial correspondence -------------------------------------------
% translation into y-axis
x0=mot.rootTranslation(1,1);
z0=mot.rootTranslation(3,1);
mot.rootTranslation(1,:)=mot.rootTranslation(1,:)-x0;
mot.rootTranslation(3,:)=mot.rootTranslation(3,:)-z0;
if (sqrt(mot.rootTranslation(1,mot.nframes)^2+mot.rootTranslation(3,mot.nframes)^2)>10)
u=[mot.rootTranslation(1,mot.nframes);mot.rootTranslation(3,mot.nframes)];
v=[1;0];
angle=acos((u'*v)/(sqrt(u'*u)));
if u(2)<0
angle=-angle;
end
else
m=quatrot([0;0;1],mot.rotationQuat{1}(:,1));
u=[m(1);m(3)];
v=[1;0];
angle=acos((u'*v)/(sqrt(u'*u)));
if u(2)<0
angle=-angle;
end
end
mot=rotateMotion(skel,mot,angle,'y');
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
translateMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/translateMotion.m
| 953 |
utf_8
|
e96682ed915c35d67d8dec51f6cf98c1
|
% function translateMotion
% translates a motion with specified translation
% mot = translateMotion(skel,mot,x,y,z)
% author: Jochen Tautges ([email protected])
function mot = translateMotion(skel,mot,x,y,z,varargin)
computetrajsbb=true;
if (nargin == 6)
computetrajsbb=varargin{1};
end
mot.rootTranslation(1,:) = mot.rootTranslation(1,:)+x;
mot.rootTranslation(2,:) = mot.rootTranslation(2,:)+y;
mot.rootTranslation(3,:) = mot.rootTranslation(3,:)+z;
if (computetrajsbb)
mot.jointTrajectories = C_forwardKinematicsQuat(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
end
if (isfield(mot,'jointVelocities')&&~isempty(mot.jointVelocities) && mot.nframes>1)
mot = addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations')&&~isempty(mot.jointAccelerations) && mot.nframes>1)
mot = addAccToMot(mot);
end
%fprintf('Motion successfully translated with x=%.2f, y=%.2f, z=%.2f.\n',x,y,z);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
removeSkating.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/removeSkating.m
| 999 |
utf_8
|
4a6011f9d697887866ed98d58c66ecea
|
% function removeSkating
% performs simple (naive) clean up of skating effects
% mot = removeSkating(skel,mot)
% author: Jochen Tautges ([email protected])
function mot = removeSkating(skel,mot)
leftJoint = 4;
rightJoint = 7;
badTranslation=0;
for i=2:mot.nframes
if mot.jointTrajectories{leftJoint}(2,i)<=mot.jointTrajectories{rightJoint}(2,i) % left foot on floor
joint = leftJoint;
else
joint = rightJoint;
end
badTranslation = badTranslation + mot.jointTrajectories{joint}([1,3],i)...
- mot.jointTrajectories{joint}([1,3],i-1);
mot.rootTranslation([1,3],i) = mot.rootTranslation([1,3],i) - badTranslation;
end
pos = cell2mat(mot.jointTrajectories([leftJoint,rightJoint]));
minY = min(min(pos(2:3:end,:)));
mot.rootTranslation(2,:) = mot.rootTranslation(2,:)-minY;
mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
removeOrientation.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/removeOrientation.m
| 744 |
utf_8
|
60ebeb2fc2ab79ed3130e0cd080802c0
|
% function removeOrientation
% removes the global orientation (root orientation) of a motion
% mot = removeOrientation(skel,mot)
% author: Jochen Tautges ([email protected])
function mot = removeOrientation(skel,mot)
if ~(isempty(mot.rotationQuat))
mot.rotationQuat{1}=[ones(1,mot.nframes);zeros(3,mot.nframes)];
end
if ~(isempty(mot.rotationEuler))
mot.rotationEuler{1}=zeros(size(mot.rotationEuler{1}));
end
mot.jointTrajectories=forwardKinematicsQuat(skel,mot);
mot.boundingBox=computeBoundingBox(mot);
if (isfield(mot,'jointVelocities')&&~isempty(mot.jointVelocities))
mot=addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations')&&~isempty(mot.jointAccelerations))
mot=addAccToMot(mot);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
addVelToMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/addVelToMot.m
| 1,798 |
utf_8
|
8972ec6857f06618cd7591854f8fd178
|
% function mot = addVelToMot(mot)
% computes joint velocities in each frame and adds the field
% 'jointVelocities' to mot structure
% additionally filters the computed velocities with binomial filter
% author: Jochen Tautges ([email protected])
function mot = addVelToMot(mot,varargin)
if nargin==2
filterSize = varargin{1};
else
filterSize = min(floor(mot.samplingRate/10),floor(mot.nframes/2));
end
if iscell(mot.jointTrajectories)
jointTrajectories = double(cell2mat(mot.jointTrajectories));
else
jointTrajectories = mot.jointTrajectories;
end
jointVelocities = zeros(size(jointTrajectories));
if mot.nframes>1
% padding
jointTrajectories = [ 3*jointTrajectories(:,1)-2*jointTrajectories(:,2),...
2*jointTrajectories(:,1)-jointTrajectories(:,2),...
jointTrajectories,...
2*jointTrajectories(:,end)-jointTrajectories(:,end-1),...
3*jointTrajectories(:,end)-2*jointTrajectories(:,end-1)];
% 5-point derivation
weights = [1 -8 0 8 -1];
for frame = 3:mot.nframes+2
jointVelocities(:,frame-2) = ...
weights(1) * jointTrajectories(:,frame-2) ...
+ weights(2) * jointTrajectories(:,frame-1) ...
+ weights(3) * jointTrajectories(:,frame) ...
+ weights(4) * jointTrajectories(:,frame+1) ...
+ weights(5) * jointTrajectories(:,frame+2);
end
jointVelocities = jointVelocities / (12 * mot.frameTime);
% filtering velocities
jointVelocities = filterTimeline(jointVelocities,filterSize,'bin');
end
mot.jointVelocities = mat2cell(jointVelocities,3*ones(1,mot.njoints),mot.nframes);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
rotateMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/rotateMotion.m
| 1,232 |
utf_8
|
d412c25a6076da06f2bf3ca4ee7d6ab7
|
% function rotateMotion
% rotates a motion with specified angle (in radians) around specified axis
% mot = rotateMotion(skel,mot,angle,axis,varargin)
% if varargin = false, jointTrajectories and boundingBox won't be computed
% author: Jochen Tautges ([email protected])
function [mot,Q] = rotateMotion(skel,mot,angle,axis,varargin)
computetrajsbb=true;
if (nargin == 5)
computetrajsbb=varargin{1};
end
Q = rotquat(angle,axis);
mot.rootTranslation = C_quatrot(mot.rootTranslation,Q);
mot.rotationQuat{1} = C_quatmult(Q,mot.rotationQuat{1});
if (computetrajsbb)
mot.jointTrajectories = C_forwardKinematicsQuat(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
end
if ~(isempty(mot.rotationEuler))
mot = convert2euler(skel,mot);
% mot.rotationEuler{1} = flipud(quat2euler(mot.rotationQuat{1},'zyx'))*180/pi;
end
if (isfield(mot,'jointVelocities') && ~isempty(mot.jointVelocities) && mot.nframes>1)
mot = addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations') && ~isempty(mot.jointAccelerations) && mot.nframes>1)
mot = addAccToMot(mot);
end
%fprintf('Motion successfully rotated with %.2f degrees around %c-axis.\n',angle*180/pi,axis);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
addBonesToMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/addBonesToMot.m
| 6,776 |
utf_8
|
31dba400ca97caa43ee8cd68534b3bb7
|
% mot = addBonesToMot(mot);
% adds field "bones" to struct "mot" with the following columns:
% - name: name of the bone (cf. struct "skel")
% - bone vectors: father to son oriented vector for each frame
% - bone length: length of the bone (cf. struct "skel")
% - normalized bone vectors
% - father and son joints related to the bone
function mot=addBonesToMot(mot)
%
% 17
% | head
% 16
% | upperneck
% 15
% | lowerneck
% lhand lradius lclavicle | rclavicle rradius rhand
% lfingers 23-22-21-20-----19------18--14--25------26-----27-28-29-30 rfingers
% lthumb | lwrist lhumerus | rhumerus rwrist | rthumb
% 24 | thorax 31
% 13
% | upperback
% |
% 12
% | lowerback
% |
% lhip 2--1--7 rhip
% / \
% / \
% lfemur / \ rfemur
% / \
% / \
% 3 8
% | |
% | |
% ltibia | | rtibia
% | |
% 4 9
% / lfoot rfoot \
% ltoes 6--5 10-11 rtoes
mot.bones.names{1,1} = 'lhip';
mot.bones.vectors{1,1} = mot.jointTrajectories{2}-mot.jointTrajectories{1};
mot.bones.joints{1,1} = [1,2];
mot.bones.names{2,1} = 'lfemur';
mot.bones.vectors{2,1} = mot.jointTrajectories{3}-mot.jointTrajectories{2};
mot.bones.joints{2,1} = [2,3];
mot.bones.names{3,1} = 'ltibia';
mot.bones.vectors{3,1} = mot.jointTrajectories{4}-mot.jointTrajectories{3};
mot.bones.joints{3,1} = [3,4];
mot.bones.names{4,1} = 'lfoot';
mot.bones.vectors{4,1} = mot.jointTrajectories{5}-mot.jointTrajectories{4};
mot.bones.joints{4,1} = [4,5];
mot.bones.names{5,1} = 'ltoes';
mot.bones.vectors{5,1} = mot.jointTrajectories{6}-mot.jointTrajectories{5};
mot.bones.joints{5,1} = [5,6];
mot.bones.names{6,1} = 'rhip';
mot.bones.vectors{6,1} = mot.jointTrajectories{7}-mot.jointTrajectories{1};
mot.bones.joints{6,1} = [1,7];
mot.bones.names{7,1} = 'rfemur';
mot.bones.vectors{7,1} = mot.jointTrajectories{8}-mot.jointTrajectories{7};
mot.bones.joints{7,1} = [7,8];
mot.bones.names{8,1} = 'rtibia';
mot.bones.vectors{8,1} = mot.jointTrajectories{9}-mot.jointTrajectories{8};
mot.bones.joints{8,1} = [8,9];
mot.bones.names{9,1} = 'rfoot';
mot.bones.vectors{9,1} = mot.jointTrajectories{10}-mot.jointTrajectories{9};
mot.bones.joints{9,1} = [9,10];
mot.bones.names{10,1} = 'rtoes';
mot.bones.vectors{10,1} = mot.jointTrajectories{11}-mot.jointTrajectories{10};
mot.bones.joints{10,1} = [10,11];
mot.bones.names{11,1} = 'lowerback';
mot.bones.vectors{11,1} = mot.jointTrajectories{12}-mot.jointTrajectories{1};
mot.bones.joints{11,1} = [1,12];
mot.bones.names{12,1} = 'upperback';
mot.bones.vectors{12,1} = mot.jointTrajectories{13}-mot.jointTrajectories{12};
mot.bones.joints{12,1} = [12,13];
mot.bones.names{13,1} = 'thorax';
mot.bones.vectors{13,1} = mot.jointTrajectories{14}-mot.jointTrajectories{13};
mot.bones.joints{13,1} = [13,14];
mot.bones.names{14,1} = 'lowerneck';
mot.bones.vectors{14,1} = mot.jointTrajectories{15}-mot.jointTrajectories{14};
mot.bones.joints{14,1} = [14,15];
mot.bones.names{15,1} = 'upperneck';
mot.bones.vectors{15,1} = mot.jointTrajectories{16}-mot.jointTrajectories{15};
mot.bones.joints{15,1} = [15,16];
mot.bones.names{16,1} = 'head';
mot.bones.vectors{16,1} = mot.jointTrajectories{17}-mot.jointTrajectories{16};
mot.bones.joints{16,1} = [16,17];
mot.bones.names{17,1} = 'lclavicle';
mot.bones.vectors{17,1} = mot.jointTrajectories{18}-mot.jointTrajectories{14};
mot.bones.joints{17,1} = [14,18];
mot.bones.names{18,1} = 'lhumerus';
mot.bones.vectors{18,1} = mot.jointTrajectories{19}-mot.jointTrajectories{18};
mot.bones.joints{18,1} = [18,19];
mot.bones.names{19,1} = 'lradius';
mot.bones.vectors{19,1} = mot.jointTrajectories{20}-mot.jointTrajectories{19};
mot.bones.joints{19,1} = [19,20];
mot.bones.names{20,1} = 'lwrist';
mot.bones.vectors{20,1} = mot.jointTrajectories{21}-mot.jointTrajectories{20};
mot.bones.joints{20,1} = [20,21];
mot.bones.names{21,1} = 'lhand';
mot.bones.vectors{21,1} = mot.jointTrajectories{22}-mot.jointTrajectories{21};
mot.bones.joints{21,1} = [21,22];
mot.bones.names{22,1} = 'lfingers';
mot.bones.vectors{22,1} = mot.jointTrajectories{23}-mot.jointTrajectories{22};
mot.bones.joints{22,1} = [22,23];
mot.bones.names{23,1} = 'lthumb';
mot.bones.vectors{23,1} = mot.jointTrajectories{24}-mot.jointTrajectories{21};
mot.bones.joints{23,1} = [21,24];
mot.bones.names{24,1} = 'rclavicle';
mot.bones.vectors{24,1} = mot.jointTrajectories{25}-mot.jointTrajectories{14};
mot.bones.joints{24,1} = [14,25];
mot.bones.names{25,1} = 'rhumerus';
mot.bones.vectors{25,1} = mot.jointTrajectories{26}-mot.jointTrajectories{25};
mot.bones.joints{25,1} = [25,26];
mot.bones.names{26,1} = 'rradius';
mot.bones.vectors{26,1} = mot.jointTrajectories{27}-mot.jointTrajectories{26};
mot.bones.joints{26,1} = [26,27];
mot.bones.names{27,1} = 'rwrist';
mot.bones.vectors{27,1} = mot.jointTrajectories{28}-mot.jointTrajectories{27};
mot.bones.joints{27,1} = [27,28];
mot.bones.names{28,1} = 'rhand';
mot.bones.vectors{28,1} = mot.jointTrajectories{29}-mot.jointTrajectories{28};
mot.bones.joints{28,1} = [28,29];
mot.bones.names{29,1} = 'rfingers';
mot.bones.vectors{29,1} = mot.jointTrajectories{30}-mot.jointTrajectories{29};
mot.bones.joints{29,1} = [29,30];
mot.bones.names{30,1} = 'rthumb';
mot.bones.vectors{30,1} = mot.jointTrajectories{31}-mot.jointTrajectories{28};
mot.bones.joints{30,1} = [28,31];
for i=1:30
mot.bones.length{i,1}=normOfColumns(mot.bones.vectors{i,1}(:,1));
end
for i=1:30
mot.bones.normalizedVectors{i,1}=mot.bones.vectors{i,1}/mot.bones.length{i,1};
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
C_rotateMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/C_rotateMotion.m
| 1,344 |
utf_8
|
84fd698ee9ee92ec250403e41ed0aa23
|
% function C_rotateMotion
% rotates a motion with specified angle (in radians) around specified axis
% mot = rotateMotion(skel,mot,angle,axis,varargin)
% if varargin = false, jointTrajectories and boundingBox won't be computed
% author: Jochen Tautges ([email protected]),
% some modifications by Tim Golla ([email protected])
function mot = C_rotateMotion(skel,mot,angle,axis,varargin)
computetrajsbb=true;
if (nargin == 5)
computetrajsbb=varargin{1};
end
Q = rotquat(angle,axis);
for i=1:mot.nframes
mot.rootTranslation(:,i) = C_quatrot(mot.rootTranslation(:,i),Q);
mot.rotationQuat{1}(:,i) = C_quatmult(Q,mot.rotationQuat{1}(:,i));
end
if (computetrajsbb)
mot.jointTrajectories = C_forwardKinematicsQuat(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
end
if ~(isempty(mot.rotationEuler))
mot = C_convert2euler(skel,mot);
% mot.rotationEuler{1} = flipud(quat2euler(mot.rotationQuat{1},'zyx'))*180/pi;
end
if (isfield(mot,'jointVelocities') && ~isempty(mot.jointVelocities) && mot.nframes>1)
mot = addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations') && ~isempty(mot.jointAccelerations) && mot.nframes>1)
mot = addAccToMot(mot);
end
%fprintf('Motion successfully rotated with %.2f degrees around %c-axis.\n',angle*180/pi,axis);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
removeTranslation.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/removeTranslation.m
| 1,318 |
utf_8
|
be462a31fcc4d7cdadf1d40f9c721fd6
|
% function removeTranslation
% removes the global translation (root translation) of a motion
% mot = removeTranslation([skel,]mot)
% if skel is specified function uses forward kinematics, otherwise simple
% subtraction of root positions from all joint positions
% author: Jochen Tautges ([email protected])
function mot = removeTranslation(varargin)
switch nargin
case 1
mot=varargin{1};
mot.jointTrajectories{1}(:,:) = 0;
for i=2:mot.njoints
mot.jointTrajectories{i} = mot.jointTrajectories{i}-mot.rootTranslation;
end
mot.rootTranslation = zeros(size(mot.rootTranslation));
case 2
if isfield(varargin{1},'jointTrajectories');
mot=varargin{1};
skel=varargin{2};
else
skel=varargin{1};
mot=varargin{2};
end
mot.rootTranslation = zeros(size(mot.rootTranslation));
mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
otherwise
error('Wrong number of argins!');
end
mot.boundingBox=computeBoundingBox(mot);
if (isfield(mot,'jointVelocities')&&~isempty(mot.jointVelocities))
mot=addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations')&&~isempty(mot.jointAccelerations))
mot=addAccToMot(mot);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
addAccToMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/motionTools/addAccToMot.m
| 1,726 |
utf_8
|
ae87460bf938a441c2cf7fd2ec602fd1
|
% function mot = addAccToMot(mot)
% computes joint accelerations in each frame and adds the field
% 'jointAccelerations' to mot structure
% additionally filters the computed accelerations with binomial filter
% author: Jochen Tautges ([email protected])
function mot = addAccToMot(mot,varargin)
if nargin==2
filterSize = varargin{1};
else
filterSize = min(floor(mot.samplingRate/10),floor(mot.nframes/2));
end
if iscell(mot.jointTrajectories)
jointTrajectories = cell2mat(mot.jointTrajectories);
else
jointTrajectories = mot.jointTrajectories;
end
jointAccelerations = zeros(size(jointTrajectories));
% padding
jointTrajectories = [ 3*jointTrajectories(:,1)-2*jointTrajectories(:,2),...
2*jointTrajectories(:,1)-jointTrajectories(:,2),...
jointTrajectories,...
2*jointTrajectories(:,end)-jointTrajectories(:,end-1),...
3*jointTrajectories(:,end)-2*jointTrajectories(:,end-1)];
% 5-point derivation
weights = [-1 16 -30 16 -1];
for frame = 3:mot.nframes+2
jointAccelerations(:,frame-2) = ...
weights(1) * jointTrajectories(:,frame-2) ...
+ weights(2) * jointTrajectories(:,frame-1) ...
+ weights(3) * jointTrajectories(:,frame) ...
+ weights(4) * jointTrajectories(:,frame+1) ...
+ weights(5) * jointTrajectories(:,frame+2);
end
jointAccelerations = jointAccelerations / (12 * mot.frameTime^2);
% filtering accelerations
jointAccelerations = filterTimeline(jointAccelerations,filterSize,'bin');
mot.jointAccelerations = mat2cell(jointAccelerations,3*ones(1,mot.njoints),mot.nframes);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
myaa.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools/plotTools/myaa.m
| 11,141 |
utf_8
|
a66dd7fc188c3f6a1a0a0c07623cf831
|
function [varargout] = myaa(varargin)
%MYAA Render figure with anti-aliasing.
% MYAA
% Anti-aliased rendering of the current figure. This makes graphics look
% a lot better than in a standard matlab figure, which is useful for
% publishing results on the web or to better see the fine details in a
% complex and cluttered plot. Some simple keyboard commands allow
% the user to set the rendering quality interactively, zoom in/out and
% re-render when needed.
%
% Usage:
% myaa: Renders an anti-aliased version of the current figure.
%
% myaa(K): Sets the supersampling factor to K. Using a
% higher K yields better rendering but takes longer time. If K is
% omitted, it defaults to 4. It may be useful to run e.g. myaa(2) to
% make a low-quality rendering as a first try, because it is a lot
% faster than the default.
%
% myaa([K D]): Sets supersampling factor to K but downsamples the
% image by a factor of D. If D is larger than K, the rendered image
% will be smaller than the original. If D is smaller than K, the
% rendering will be bigger.
%
% myaa('publish'): An experimental parameter, useful for publishing
% matlab programs (see example 3). Beware, it kills the current figure.
%
% Interactivity:
% The anti-aliased figure can be updated with the following keyboard
% commands:
%
% <space> Re-render image (to reflect changes in the figure)
% + Zoom in (decrease downsampling factor)
% - Zoom out (increase downsampling factor)
% 1 ... 9 Change supersampling and downsampling factor to ...
% q Quit, i.e. close the anti-aliased figure
%
% Myaa can also be called with up to 3 parameters.
% FIG = MYAA(K,AAMETHOD,FIGMODE)
% Parameters and output:
% K Subsampling factor. If a vector is specified, [K D], then
% the second element will describe the downsampling factor.
% Default is K = 4 and D = 4.
% AAMETHOD Downsampling method. Normally this is chosen automatically.
% 'standard': convolution based filtering and downsampling
% 'imresize': downsampling using the imresize command from
% the image toolbox.
% 'noshrink': used internally
% FIGMODE Display mode
% 'figure': the normal mode, a new figure is created
% 'update': internally used for interactive sessions
% 'publish': used internally
% FIG A handle to the new anti-aliased figure
%
% Example 1:
% spharm2;
% myaa;
% % Press '1', '2' or '4' and try '+' and '-'
% % Press 'r' or <space> to update the anti-aliased rendering, e.g.
% % after rotating the 3-D object in the original figure.
%
% Example 2:
% line(randn(2500,2)',randn(2500,2)','color','black','linewidth',0.01)
% myaa(8);
%
% Example 3:
% xpklein;
% myaa(2,'standard');
%
% Example 3:
% Put the following in test.m
% %% My test publish
% % Testing to publish some anti-aliased images
% %
% spharm2; % Produce some nice graphics
% myaa('publish'); % Render an anti-aliased version
%
% Then run:
% publish test.m;
% showdemo test;
%
%
% BUGS:
% Dotted and dashed lines in plots are not rendered correctly. This is
% probably due to a bug in Matlab and it will hopefully be fixed in a
% future version.
% The OpenGL renderer does not always manage to render an image large
% enough. Try the zbuffer renderer if you have problems or decrease the
% K factor. You can set the current renderer to zbuffer by running e.g.
% set(gcf,'renderer','zbuffer').
%
% See also PUBLISH, PRINT
%
% Version 1.1, 2008-08-21
% Version 1.0, 2008-08-05
%
% Author: Anders Brun
% [email protected]
%
%% Force drawing of graphics
drawnow;
%% Find out about the current DPI...
screen_DPI = get(0,'ScreenPixelsPerInch');
%% Determine the best choice of convolver.
% If IPPL is available, imfilter is much faster. Otherwise it does not
% matter too much.
try
if ippl()
myconv = @imfilter;
else
myconv = @conv2;
end
catch
myconv = @conv2;
end
%% Set default options and interpret arguments
if isempty(varargin)
self.K = [4 4];
try
imfilter(zeros(2,2),zeros(2,2));
self.aamethod = 'imresize';
catch
self.aamethod = 'standard';
end
self.figmode = 'figure';
elseif strcmp(varargin{1},'publish')
self.K = [4 4];
self.aamethod = 'noshrink';
self.figmode = 'publish';
elseif strcmp(varargin{1},'update')
self = get(gcf,'UserData');
figure(self.source_fig);
drawnow;
self.figmode = 'update';
elseif strcmp(varargin{1},'lazyupdate')
self = get(gcf,'UserData');
self.figmode = 'lazyupdate';
elseif length(varargin) == 1
self.K = varargin{1};
if length(self.K) == 1
self.K = [self.K self.K];
end
if self.K(1) > 16
error('To avoid excessive use of memory, K has been limited to max 16. Change the code to fix this on your own risk.');
end
try
imfilter(zeros(2,2),zeros(2,2));
self.aamethod = 'imresize';
catch
self.aamethod = 'standard';
end
self.figmode = 'figure';
elseif length(varargin) == 2
self.K = varargin{1};
self.aamethod = varargin{2};
self.figmode = 'figure';
elseif length(varargin) == 3
self.K = varargin{1};
self.aamethod = varargin{2};
self.figmode = varargin{3};
if strcmp(self.figmode,'publish') && ~strcmp(varargin{2},'noshrink')
printf('\nThe AAMETHOD was not set to ''noshrink'': Fixed.\n\n');
self.aamethod = 'noshrink';
end
else
error('Wrong syntax, run: help myaa');
end
if length(self.K) == 1
self.K = [self.K self.K];
end
%% Capture current figure in high resolution
if ~strcmp(self.figmode,'lazyupdate');
tempfile = 'myaa_temp_screendump.png';
self.source_fig = gcf;
current_paperpositionmode = get(self.source_fig,'PaperPositionMode');
current_inverthardcopy = get(self.source_fig,'InvertHardcopy');
set(self.source_fig,'PaperPositionMode','auto');
set(self.source_fig,'InvertHardcopy','off');
print(self.source_fig,['-r',num2str(screen_DPI*self.K(1))], '-dpng', tempfile);
set(self.source_fig,'InvertHardcopy',current_inverthardcopy);
set(self.source_fig,'PaperPositionMode',current_paperpositionmode);
self.raw_hires = imread(tempfile);
delete(tempfile);
end
%% Start filtering to remove aliasing
w = warning;
warning off;
if strcmp(self.aamethod,'standard') || strcmp(self.aamethod,'noshrink')
% Subsample hires figure image with standard anti-aliasing using a
% butterworth filter
kk = lpfilter(self.K(2)*3,self.K(2)*0.9,2);
mm = myconv(ones(size(self.raw_hires(:,:,1))),kk,'same');
a1 = max(min(myconv(single(self.raw_hires(:,:,1))/(256),kk,'same'),1),0)./mm;
a2 = max(min(myconv(single(self.raw_hires(:,:,2))/(256),kk,'same'),1),0)./mm;
a3 = max(min(myconv(single(self.raw_hires(:,:,3))/(256),kk,'same'),1),0)./mm;
if strcmp(self.aamethod,'standard')
if abs(1-self.K(2)) > 0.001
raw_lowres = double(cat(3,a1(2:self.K(2):end,2:self.K(2):end),a2(2:self.K(2):end,2:self.K(2):end),a3(2:self.K(2):end,2:self.K(2):end)));
else
raw_lowres = self.raw_hires;
end
else
raw_lowres = double(cat(3,a1,a2,a3));
end
elseif strcmp(self.aamethod,'imresize')
% This is probably the fastest method available at this moment...
raw_lowres = single(imresize(self.raw_hires,1/self.K(2),'bilinear'))/256;
end
warning(w);
%% Place the anti-aliased image in some image on the screen ...
if strcmp(self.figmode,'figure');
% Create a new figure at the same place as the previous
% The content of this new image is just a bitmap...
oldpos = get(gcf,'Position');
self.myaa_figure = figure;
fig = self.myaa_figure;
set(fig,'Menubar','none');
set(fig,'Resize','off');
sz = size(raw_lowres);
set(fig,'Units','pixels');
pos = [oldpos(1:2) sz(2:-1:1)];
set(fig,'Position',pos);
ax = axes;
image(raw_lowres);
set(ax,'Units','pixels');
set(ax,'Position',[1 1 sz(2) sz(1)]);
axis off;
elseif strcmp(self.figmode,'publish');
% Create a new figure at the same place as the previous
% The content of this new image is just a bitmap...
self.myaa_figure = figure;
fig = self.myaa_figure;
current_units = get(self.source_fig,'Units');
set(self.source_fig,'Units','pixels');
pos = get(self.source_fig,'Position');
set(self.source_fig,'Units',current_units);
set(fig,'Position',[pos(1) pos(2) pos(3) pos(4)]);
ax = axes;
image(raw_lowres);
set(ax,'Units','normalized');
set(ax,'Position',[0 0 1 1]);
axis off;
close(self.source_fig);
elseif strcmp(self.figmode,'update');
fig = self.myaa_figure;
figure(fig);
clf;
set(fig,'Menubar','none');
set(fig,'Resize','off');
sz = size(raw_lowres);
set(fig,'Units','pixels');
pos = get(fig,'Position');
pos(3:4) = sz(2:-1:1);
set(fig,'Position',pos);
ax = axes;
image(raw_lowres);
set(ax,'Units','pixels');
set(ax,'Position',[1 1 sz(2) sz(1)]);
axis off;
elseif strcmp(self.figmode,'lazyupdate');
clf;
fig = self.myaa_figure;
sz = size(raw_lowres);
pos = get(fig,'Position');
pos(3:4) = sz(2:-1:1);
set(fig,'Position',pos);
ax = axes;
image(raw_lowres);
set(ax,'Units','pixels');
set(ax,'Position',[1 1 sz(2) sz(1)]);
axis off;
end
%% Store current state
set(gcf,'userdata',self);
set(gcf,'KeyPressFcn',@keypress);
set(gcf,'Interruptible','off');
%% Avoid unnecessary console output
if nargout == 1
varargout(1) = {fig};
end
%% A simple lowpass filter kernel (Butterworth).
% sz is the size of the filter
% subsmp is the downsampling factor to be used later
% n is the degree of the butterworth filter
function kk = lpfilter(sz, subsmp, n)
sz = 2*floor(sz/2)+1; % make sure the size of the filter is odd
cut_frequency = 0.5 / subsmp;
range = (-(sz-1)/2:(sz-1)/2)/(sz-1);
[ii,jj] = ndgrid(range,range);
rr = sqrt(ii.^2+jj.^2);
kk = ifftshift(1./(1+(rr./cut_frequency).^(2*n)));
kk = fftshift(real(ifft2(kk)));
kk = kk./sum(kk(:));
function keypress(src,evnt)
if isempty(evnt.Character)
return
end
recognized = 0;
self = get(gcf,'userdata');
if evnt.Character == '+'
self.K(2) = max(self.K(2).*0.5^(1/2),1);
recognized = 1;
set(gcf,'userdata',self);
myaa('lazyupdate');
elseif evnt.Character == '-'
self.K(2) = min(self.K(2).*2^(1/2),16);
recognized = 1;
set(gcf,'userdata',self);
myaa('lazyupdate');
elseif evnt.Character == ' ' || evnt.Character == 'r' || evnt.Character == 'R'
set(gcf,'userdata',self);
myaa('update');
elseif evnt.Character == 'q'
close(gcf);
elseif find('123456789' == evnt.Character)
self.K = [str2double(evnt.Character) str2double(evnt.Character)];
set(gcf,'userdata',self);
myaa('update');
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
H_writeImageTest2.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/H_writeImageTest2.m
| 1,356 |
utf_8
|
900cb28147fd72b40923c1c56db8fdf0
|
function H_writeImageTest2(opts,inp2d,est2dRt,vFile,f,bodyType)
ifr = opts.sFrame + f - 1;
if(isobject(vFile))
rgbIm = read(vFile,ifr);
else
imn = [vFile opts.subject '_' opts.action '_' num2str(opts.frameNo(1,f)) '_' num2str(opts.frameNo(2,f)) '.png'];
rgbIm = imread(imn);
end
clf
imshow(rgbIm,'Border','tight');
%%
xgt = inp2d(1:2:end,f);
ygt = inp2d(2:2:end,f);
xObj = est2dRt(1:2:end,f); % pose
yObj = est2dRt(2:2:end,f);
hold on
%%
H_drawLine(xgt,ygt);
plot(xgt,ygt,'o','MarkerSize',6,'MarkerEdgeColor','r','MarkerFaceColor','y','LineWidth',2);
%plot(xgt(13),ygt(13),'o','MarkerSize',6,'MarkerEdgeColor','r','MarkerFaceColor','r','LineWidth',2);
plot(xObj,yObj,'*','MarkerSize',4,'MarkerEdgeColor','g');
xm = get(gca,'xLim');
ym = get(gca,'yLim');
text (xm(1) + 20,ym(1) + 20,['Frame # ' num2str(ifr)],'Color','y', 'FontWeight','bold', 'FontSize',15);
%text (xm(1) + 20,ym(1) + 40,['Count # ' num2str(f)],'Color','y', 'FontWeight','bold', 'FontSize',15);
pause(0.4);
%%
imPath = [opts.saveResPath 'optimTestIm\'];
if ~exist(imPath,'dir')
mkdir(imPath);
end
imName = [imPath opts.subject '_' opts.actName '_' num2str(opts.knn) '_' bodyType '_' num2str(ifr) '_' num2str(f) '.png'];
frm = getframe(gcf);
im = frame2im(frm);
imwrite(im, imName);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
H_text2MotFile.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/H_text2MotFile.m
| 3,893 |
utf_8
|
0d21eef1f64aae54f67fb40229623cc7
|
function H_text2MotFile(dataName, strPath,dimInfo)
motIn = emptyMotionLocal;
strName = [strPath dataName];
fid = fopen(strName);
cSpace = char('\n');
txtData = fscanf(fid, ['%f', cSpace]);
token = textscan(dataName, '%s', 'delimiter', '_');
motIn.njoints = 14;
frmInfo = 2; % video data and mocap data
incr = motIn.njoints * dimInfo + frmInfo;
if(mod(length(txtData),incr))
txtData(1) = [];
end
motIn.nframes = length(txtData)/incr;
txtData = reshape(txtData, incr, length(txtData)/incr);
i = 3;
for k = 1:motIn.njoints
motIn.jointTrajectories{k,1}(:,:) = txtData(i:i+2,:);
i = i + 3;
end
%%
% mot.jointNames = {'Right Ankle'; 'Right Knee'; 'Right Hip'; 'Left Hip'; 'Left Knee'; 'Left Ankle';...
% 'Right Wrist'; 'Right Elbow'; 'Right Shoulder'; 'Left Shoulder';'Left Elbow';...
% 'Left Wrist'; 'Neck'; 'Head'};
motIn.jointNames = {'Left Ankle'; 'Left Knee'; 'Left Hip'; 'Right Hip'; 'Right Knee'; 'Right Ankle';...
'Left Wrist'; 'Left Elbow'; 'Left Shoulder'; 'Right Shoulder';'Right Elbow';...
'Right Wrist'; 'Neck'; 'Head'};
motIn.markerNames = {'RTIO'; 'RFEO'; 'RFEP'; 'LFEP'; 'LFEO'; 'LTIO'; 'RRAO'; 'RHUO'; ...
'RCLO'; 'LCLO'; 'LHUO'; 'LRAO'; 'TRXO'; 'LFHD'; 'RFHD'};
motIn.samplingRate = 120;
motIn.frameTime = 1/motIn.samplingRate;
motIn.subject = token{1,1}{1,1};
motIn.action = token{1,1}{2,1};
motIn.trail = token{1,1}{3,1};
motIn.camera = strtok(token{1,1}{4,1}, '.');
motIn.filename = dataName;
motIn.frameNumbers(1,:) = txtData(1,:);
motIn.frameNumbers(2,:) = txtData(2,:);
motIn.vidStartFrame = txtData(1);
motIn.vidEndFrame = txtData(1,motIn.nframes);
motIn.mocStartFrame = txtData(2);
motIn.mocEndFrame = txtData(2,motIn.nframes);
if(motIn.vidStartFrame == motIn.vidEndFrame)
motIn.vidEndFrame = motIn.vidStartFrame + motIn.nframes - 1;
end
if (dimInfo == 2)
motIn.dimData = '2D';
else
motIn.dimData = '3D';
end
motIn.boundingBox = computeBoundingBox(motIn);
for i = 1:length(motIn.markerNames)
motIn.nameMap{i,1} = char(motIn.markerNames(i));
motIn.nameMap{i,2} = 0;
motIn.nameMap{i,3} = i;
end
%% filp the left side towards right side
% qy = rotquat(180,'y');
% for m = 1:mot.njoints
% mot.jointTrajectories{m}(:,:) = quatrot(mot.jointTrajectories{m}(:,:),qy);
% end
%%
strRes = strtok(dataName, '.');
save(fullfile(strPath,[strRes '.mat']),'motIn','-v7.3');
end
function mot = emptyMotionLocal
%%
mot = struct('name','pgIlya',...
'njoints',0,... % number of joints
'nframes',0,... % number of frames
'frameTime',nan,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',nan,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',cell(1,1),... % 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',cell(1,1),... % cell array of joint names: maps node ID to joint name
'markerNames',cell(1,1),... % cell array of marker names:
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename',''); % source filename
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
optimCamMtxPerspSepWinH30Mbm.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/optimCamMtxPerspSepWinH30Mbm.m
| 5,937 |
utf_8
|
71e484e9935760067294b595c9b74540
|
function [est2dRt, est3dRt, X] = optimCamMtxPerspSepWinH30Mbm(inp2d,knn3d,cm,opts, singleDim, varargin)
if(nargin > 5)
vFile = varargin{1};
end
if (exist('vFile','var'))
if(~isobject(vFile))
vFilePath = vFile;
end
testIm = 1;
dirFiles = dir([vFile '*.png']);
drnm = char(dirFiles.name); % orginizing order
drnmCell = cellstr(drnm);
drnmCell = sort_nat(drnmCell);
else
testIm = 0;
end
windowsize = 2; % for 5 we have to select 4
optimOpts = optimset(...
'Display','off',... % 'Display','iter'
'MaxIter',10000,...
'MaxFunEvals',10000,... % 'PlotFcns', @optimplotx, ...
'Algorithm' ,'trust-region-reflective',... %'levenberg-marquardt' , ...
'TolX',0.001);
lb = []; ub = [];
%% Loading Camera matx already optimized and save previously
pload = '../Datatest/';
camLoad = [pload opts.subject '_' opts.actName '_' num2str(opts.knn) '_pose_camMtxdfgff.mat'];
if(exist(camLoad,'file'))
load(camLoad);
end
if(exist('camMtx','var'))
camMtxLoad = true;
else
camMtxLoad = false;
startValue = [0;0;0;0;0;0];
end
%%
%--------------------------------------------------------------------------%Loading cam mtx given
cam = 2; % cam = 1 optimal
Sequence = H36MSequence(11, 1, 1, cam, 1);
obj = Sequence.getCamera();
cm.R = obj.R;
cm.T = obj.T;
cm = obj;
%%
nJoints = length(opts.allJoints);
knn3dv = knn3d(opts.cJidx3,:,:);
inp2dv = inp2d(opts.cJidx2,:);
X = cell(size(inp2d,2),1);
rn = nan(size(inp2d,2),1);
% X2 = cell(size(inp2d,2),1);
% rn2 = nan(size(inp2d,2),1);
act = 1;
if((ndims(knn3d)==3) || (singleDim == 1))
%% all knn simultaneously
est2dRt = nan(2*nJoints*opts.knn,size(inp2d,2));
est3dRt = nan(3*nJoints*opts.knn,size(inp2d,2));
fprintf('Optimization: Frame no. ... ');
count = 1;
tic
rot = true;
for f = 1:size(inp2d,2)
fprintf('%5d ', f);
%------------------------------------------------------------------% all 14 joints
est3d = H_regularize(knn3d,f);
%------------------------------------------------------------------% optimize with selective joints (opts.cJoints)
est3dv = H_regularize(knn3dv,f);
inp2dRev = reshape(inp2dv(:,f),2,[]);
inp2dRev = repmat(inp2dRev,1,size(knn3d,2));
%------------------------------------------------------------------% optimize and getting R,t parameters
%startValue0 = [0;0;0;0;0;0];
if(camMtxLoad)
startValue = camMtx{f};
end
% % % if(act<opts.act(f))
% % % startValue = [0;0;0;0;0;0];
% % % end
[X{count,1},rn(count,1)] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),startValue,lb,ub,optimOpts);
%startValue = X{count,1};
% % % if(startValue ~= [0;0;0;0;0;0])
% % % if(rn(count,1) > 9.0e5)
% % % startValue = [0;0;0;0;0;0];
% % % [x2,rnn2] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),startValue,lb,ub,optimOpts);
% % % if(rn(count,1) > rnn2)
% % % X{count,1} = x2;
% % % rn(count,1)= rnn2;
% % % end
% % % end
% % % end
% % % startValue = X{count,1};
% if(camMtxLoad)
% startValue = camMtx{f};
% [X2{count,1},rn2(count,1)] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),startValue,lb,ub,optimOpts);
% end
% if(rn2(count,1) < rn(count,1))
% Xm = X2{count,1};
% else
% Xm = X{count,1};
% end
Xm = X{count,1};
%------------------------------------------------------------------
[est2d,j3d] = H_proj(Xm,est3d,cm);
tm2d = reshape(est2d,2*nJoints,[]);
est2dRt(:,f) = tm2d(:);
%est3d(1,:) = -est3d(1,:);
j3d([2 3],:) = flipud(j3d([2 3],:));
tm3d = reshape(j3d,3*nJoints,[]);
est3dRt(:,f) = tm3d(:);
camMtx{count,1} = Xm;
fprintf('\b\b\b\b\b\b');
%%
if(testIm)
bodyType = opts.bodyType;
if(~isobject(vFile))
vFile = [vFilePath drnmCell{f}];
end
H_writeImageTest(opts,inp2d(:,f),est2dRt(:,f),vFile,f,bodyType)
end
%%
count = count + 1;
%act = opts.act(f);
end
toc
end
camName = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' opts.bodyType '_camMtx.mat'];
save(fullfile(opts.saveResPath,camName),'camMtx','-v7.3');
end
function f = objfunLocal(x,in2d,est3d,cm)
[est2d,~] = H_proj(x,est3d,cm);
f = sqrt(sum((est2d - in2d).^2));
%f = norm(est2d - in2d,1);
% f = repmat(sum(sqrt(sqrt(sum((est2d - in2d).^2)))),size(x));
end
function [est2d,j3d] = H_proj(x,est3d,cm)
R = makehgtform('xrotate',deg2rad(x(1)),'yrotate',deg2rad(x(2)),'zrotate',deg2rad(x(3)));
R = R(1:3,1:3);
T = [x(4); x(5); x(6)];
R = [R T];
j3d = R*est3d;
% T = T';
% j3d = est3d(1:3,:);
[est2d, ~] = ProjectPointRadial(j3d', cm.R, cm.T, cm.f, cm.c, cm.k, cm.p);
est2d = est2d';
end
function est3d = H_rotFlip(est3d,rot)
if(rot)
%R = makehgtform('xrotate',H_deg2rad(180));
R = makehgtform('xrotate',deg2rad(90));
est3d = R * est3d;
end
end
function est3d = H_regularize(knn3d,f)
%%
est3d = knn3d(:,:,f);
est3d = squeeze(est3d);
est3d = reshape(est3d,3,[]);
est3d([2 3],:) = flipud(est3d([2 3],:));
% est3d(2,:) = -est3d(2,:);
est3d(1,:) = -est3d(1,:); % used for h36m
est3d(4,:) = 1;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
optimCamMtxPerspSepWin.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/optimCamMtxPerspSepWin.m
| 6,857 |
utf_8
|
9f8e6572075ba03fd71b8a7a17dcc258
|
function [est2dRt, est3dRt, X] = optimCamMtxPerspSepWin(inp2d,knn3d,cm,opts,varargin)
%% Hashim Yasin
strtOp = 1;
windowsize = 4; % for 5 we have to select 4
if(nargin > 4)
vFile = varargin{1};
end
if (exist('vFile','var'))
testIm = 1;
else
testIm = 0;
end
optimOpts = optimset(...
'Display','off',... % 'Display','iter'
'MaxIter',10000,...
'MaxFunEvals',10000,... % 'PlotFcns', @optimplotx, ...
'Algorithm' , 'trust-region-reflective' , ...%'trust-region-reflective',... %'levenberg-marquardt' , ...
'TolX',0.001);
startValue = [0;0;0;0;0;0];
lb = []; ub = [];
nJoints = length(opts.allJoints);
knn3dv = knn3d(opts.cJidx3,:,:);
inp2dv = inp2d(opts.cJidx2,:);
% idx2 = H_getSingleJntIdx({'Head';'Left Shoulder'; 'Right Shoulder'; 'Left Hip'; 'Right Hip'; 'Neck'},2,opts.allJoints);
% idx3 = H_getSingleJntIdx({'Head';'Left Shoulder'; 'Right Shoulder'; 'Left Hip'; 'Right Hip'; 'Neck'},3,opts.allJoints);
% knn3dv = knn3d(idx3,:,:);
% inp2dv = inp2d(idx2,:);
cnt = 0;
if(ndims(knn3d)==3)
%% all knn simultaneously
%---------------------------------------------------------------------- starting options
if(strtOp)
fprintf('Initialization Step: ... ');
c = 1;
for s = 1:10
optimOpts2 = optimset(...
'Display','off',... % 'Display','iter'
'MaxIter',10000,...
'MaxFunEvals',10000,... % 'PlotFcns', @optimplotx, ...
'Algorithm' , 'levenberg-marquardt' , ...%'trust-region-reflective',... %'levenberg-marquardt' , ...
'TolX',0.001);
lb2 = [0,0,0,-10000,-10000,-10000];
ub2 = [360,360,360,10000,10000,10000];
est3dv = H_regularize(knn3d,s);
inp2dRev = reshape(inp2d(:,s),2,[]);
inp2dRev = repmat(inp2dRev,1,size(knn3d,2));
[Xt{c,1},rnt(c,1)] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),startValue,lb2,ub2,optimOpts2);
startValue = Xt{c,1};
c = c + 1;
end
[~,ind] = min(rnt(:));
startValue = Xt{ind,1};
end
%----------------------------------------------------------------------
est2dRt = nan(2*nJoints*opts.knn,size(inp2d,2));
est3dRt = nan(3*nJoints*opts.knn,size(inp2d,2));
fprintf('Optimization: Frame no. ... ');
tic
for f = 1:size(inp2d,2)
fprintf('%3d ', f);
%------------------------------------------------------------------% all 14 joints
est3d = H_regularize(knn3d,f);
% inp2dRe = reshape(inp2d(:,f),2,[]);,
% inp2dRe = repmat(inp2dRe,1,size(knn3d,2));
%------------------------------------------------------------------% optimize with selective joints (opts.cJoints)
est3dv = H_regularize(knn3dv,f);
inp2dRev = reshape(inp2dv(:,f),2,[]);
inp2dRev = repmat(inp2dRev,1,size(knn3d,2));
%------------------------------------------------------------------% optimize and getting R,t parameters
[X{f,1},rn(f,1)] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),startValue,lb,ub,optimOpts);
%------------------------------------------------------------------% starting options
if(f == 1)
rnFr = 0;
while(rn>3e+05)
[X{f,1},rn(f,1)] = lsqnonlin(@(x) objfunLocal(x,inp2dRev,est3dv,cm),X{f,1},lb,ub,optimOpts);
if (rnFr == rn)
break;
end
rnFr = rn;
cnt = cnt + 1;
end
end
%------------------------------------------------------------------
%R = makehgtform('xrotate',X{f,1}(1),'yrotate',X{f,1}(2),'zrotate',X{f,1}(3));
R = makehgtform('xrotate',deg2rad(X{f,1}(1)),'yrotate',deg2rad(X{f,1}(2)),'zrotate',deg2rad(X{f,1}(3)));
R = R(1:3,1:3);
T = [X{f,1}(4); X{f,1}(5); X{f,1}(6)];
R = [R T];
camMtx{f,1} = R;
%------------------------------------------------------------------% getting 2D data
j3d = R*est3d;
est2d = project_points2Short(j3d,cm.omc_ext,cm.Tc_ext,cm.fc,cm.cc,cm.kc,cm.alpha_c);
tm2d = reshape(est2d,2*nJoints,[]);
est2dRt(:,f) = tm2d(:);
%------------------------------------------------------------------% getting 3D data
j3d([2 3],:) = flipud(j3d([2 3],:)); % '123' ---> '132' here back fliping is done in opposite direction
j3d([1 3],:) = flipud(j3d([1 3],:)); % '321' ---> '123'
tm3d = reshape(j3d,3*nJoints,[]);
est3dRt(:,f) = tm3d(:);
%------------------------------------------------------------------% startValue with window size 5
sf = max(1,f-windowsize);
[~,ind] = min(rn(sf:f));
ind = ind + sf - 1;
startValue = X{ind,1};
fprintf('\b\b\b\b');
%%
if(testIm)
%%% H_writeImageTest2(opts,inp2d,est2dRt,vFile,f,bodyType)
end
%%
est3d = [];
est3dv = [];
end
end
toc
orName = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' opts.bodyType '_rn.mat'];
save(fullfile(opts.saveResPath,orName),'rn','-v7.3');
rtName = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' opts.bodyType '_rtMtx.mat'];
save(fullfile(opts.saveResPath,rtName),'X','-v7.3');
camName = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' opts.bodyType '_camMtxComb.mat'];
save(fullfile(opts.saveResPath,camName),'camMtx','-v7.3');
end
function f = objfunLocal(x,in2d,est3d,cm)
%R = makehgtform('xrotate',x(1),'yrotate',x(2),'zrotate',x(3));
R = makehgtform('xrotate',deg2rad(x(1)),'yrotate',deg2rad(x(2)),'zrotate',deg2rad(x(3)));
R = R(1:3,1:3);
T = [x(4); x(5); x(6)];
R = [R T];
j3d = R*est3d;
%%est2d = project_points2(j3d,cm.omc_ext,cm.Tc_ext,cm.fc,cm.cc,zeros(5,1),0);
%%est2d = project_points2Short(j3d,cm.omc_ext,cm.Tc_ext,cm.fc,cm.cc,cm.kc,cm.alpha_c);
est2d = project_points2ShortOptim(j3d,cm.omc_ext,cm.Tc_ext,cm.fc,cm.cc); %
f = sqrt(sum((est2d - in2d).^2));
% f = repmat(sum(sqrt(sqrt(sum((est2d - in2d).^2)))),size(x));
end
function est3d = H_regularize(knn3d,f)
%%
est3d = knn3d(:,:,f);
est3d = squeeze(est3d);
est3d = reshape(est3d,3,[]);
est3d ([1 3],:) = flipud(est3d([1 3],:)); % '321' ---> '123'
est3d ([2 3],:) = flipud(est3d([2 3],:)); % '123' ---> '132'
est3d(4,:) = 1;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
errorHumanEva.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/errorHumanEva.m
| 3,107 |
utf_8
|
274fbb095679590e793911c138288956
|
function [errPrJnts, errPrFr, errPrAllFr] = errorHumanEva(data,mocapGT,cJoints)
% data = data(1:3*length(cJoints),:);
% mocapGT = mocapGT(1:3*length(cJoints),:);
errPrFr = nan(size(data,2),1); % error nrOfFrames
errPrJnts = nan(size(data,1)/3,1,size(data,2)); % error joints*knn*nrOfFrames
for f = 1:size(data,2) % frames
errPrJntsTmp = nan(size(data,1)/3,1);
i = 1;
for j = 1:size(data,1)/3 % joints
errPrJntsTmp(j) = error_internal(data(i:i+2,f),mocapGT(i:i+2,f));
i = i + 3;
end
errPrJnts(:,1,f) = errPrJntsTmp;
errPrFr(f) = mean(errPrJntsTmp);
end
errPrAllFr = mean(errPrFr);
end
function [err] = error_internal(point1, point2)
err = sqrt(sum((point1(:) - point2(:)).^2));
end
%%
% % % % Thorax joint
% % % e1 = error_internal(pose1.torsoProximal, pose2.torsoProximal);
% % % e2 = error_internal(pose1.headProximal, pose2.headProximal);
% % % if (any([e1, e2]))
% % % err(end+1) = mean([e1, e2]);
% % % end
% % % % Pelvis joint
% % % err(end+1) = error_internal(pose1.torsoDistal, pose2.torsoDistal);
% % % % Left shoulder
% % % err(end+1) = error_internal(pose1.upperLArmProximal, pose2.upperLArmProximal);
% % % % Left elbow
% % % e1 = error_internal(pose1.upperLArmDistal, pose2.upperLArmDistal);
% % % e2 = error_internal(pose1.lowerLArmProximal, pose2.lowerLArmProximal);
% % % if (any([e1, e2]))
% % % err(end+1) = mean([e1, e2]);
% % % end
% % % % Left wrist
% % % err(end+1) = error_internal(pose1.lowerLArmDistal, pose2.lowerLArmDistal);
% % % % Right shoulder
% % % err(end+1) = error_internal(pose1.upperRArmProximal, pose2.upperRArmProximal);
% % % % Right elbow
% % % e1 = error_internal(pose1.upperRArmDistal, pose2.upperRArmDistal);
% % % e2 = error_internal(pose1.lowerRArmProximal, pose2.lowerRArmProximal);
% % % if (any([e1, e2]))
% % % err(end+1) = mean([e1, e2]);
% % % end
% % % % Right wrist
% % % err(end+1) = error_internal(pose1.lowerRArmDistal, pose2.lowerRArmDistal);
% % % % Left hip
% % % err(end+1) = error_internal(pose1.upperLLegProximal, pose2.upperLLegProximal);
% % % % Left knee
% % % e1 = error_internal(pose1.upperLLegDistal, pose2.upperLLegDistal);
% % % e2 = error_internal(pose1.lowerLLegProximal, pose2.lowerLLegProximal);
% % % if (any([e1, e2]))
% % % err(end+1) = mean([e1, e2]);
% % % end
% % % % Left ankle
% % % err(end+1) = error_internal(pose1.lowerLLegDistal, pose2.lowerLLegDistal);
% % % % Right hip
% % % err(end+1) = error_internal(pose1.upperRLegProximal, pose2.upperRLegProximal);
% % % % Right knee
% % % e1 = error_internal(pose1.upperRLegDistal, pose2.upperRLegDistal);
% % % e2 = error_internal(pose1.lowerRLegProximal, pose2.lowerRLegProximal);
% % % if (any([e1, e2]))
% % % err(end+1) = mean([e1, e2]);
% % % end
% % % % Right ankle
% % % err(end+1) = error_internal(pose1.lowerRLegDistal, pose2.lowerRLegDistal);
% % % % Head (top of the head)
% % % err(end+1) = error_internal(pose1.headDistal, pose2.headDistal);
% % % err = mean(err);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
emptyMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/emptyMotion.m
| 6,889 |
utf_8
|
6ecb50721bd9279da049272df1df4952
|
function mot = emptyMotion(varargin)
switch nargin
case 0
mot = struct('njoints',0,... % number of joints
'nframes',0,... % number of frames
'frameTime',nan,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',nan,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',cell(1,1),... % 3D joint trajectories
'jointTrajectoriesN',cell(1,1),... % 3D joint trajectories (Normalized)
'jointTrajectories2D',cell(1,1),... % 2D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',cell(1,1),... % rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',cell(1,1),... % cell array of joint names: maps node ID to joint name
'boneNames',cell(1,1),... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',cell(1,1),... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',[],... % vector of IDs for animated joints/bones
'unanimated',[],... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
case 1
if ismot_local(varargin{1})
refmot = varargin{1};
mot = struct('njoints',refmot.njoints,... % number of joints
'nframes',0,... % number of frames
'frameTime',refmot.frameTime,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',refmot.samplingRate,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',[],...cell(refmot.njoints,1),...% 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',[],...{cell(refmot.njoints,1)},...% rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',[],...{cell(refmot.njoints,1)},...% rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',{refmot.jointNames},... % cell array of joint names: maps node ID to joint name
'boneNames',{refmot.boneNames},... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',{refmot.nameMap},... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',refmot.animated,... % vector of IDs for animated joints/bones
'unanimated',refmot.unanimated,... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
elseif isskel_local(varargin{1})
skel = varargin{1};
mot = struct('njoints',skel.njoints,... % number of joints
'nframes',0,... % number of frames
'frameTime',nan,... % inverse sampling rate: time per frame (in seconds)
'samplingRate',nan,... % sampling rate (in Hertz) (120 Hertz is Carnegie-Mellon Mocap-DB standard)
'jointTrajectories',[],...{cell(skel.njoints,1)},...% 3D joint trajectories
'rootTranslation',[],... % global translation data stream of the root
'rotationEuler',[],...{cell(skel.njoints,1)},... % rotational data streams for all joints, including absolute root rotation at pos. 1, Euler angles
'rotationQuat',[],...{cell(skel.njoints,1)},... % rotational data streams for all joints, including absolute root rotation at pos. 1, quaternions
'jointNames',{skel.jointNames},... % cell array of joint names: maps node ID to joint name
'boneNames',{skel.boneNames},... % cell array of bone names: maps bone ID to node name. ID 1 is the root.
'nameMap',{skel.nameMap},... % cell array mapping standard joint names to DOF IDs and trajectory IDs
'animated',skel.animated,... % vector of IDs for animated joints/bones
'unanimated',skel.unanimated,... % vector of IDs for unanimated joints/bones
'boundingBox',[],... % bounding box (given a specific skeleton)
'filename','',... % source filename
'documentation','',... % documentation from source file
'angleUnit','deg'); % angle unit, either deg or rad
end
end
end
%% local functions
function ismot_bool = ismot_local(arg)
if isfield(arg,'nframes') && isfield(arg,'rotationQuat')
ismot_bool = true;
else
ismot_bool = false;
end
end
function isskel_bool = isskel_local(arg)
if isfield(arg,'njoints') && isfield(arg,'nodes')
isskel_bool = true;
else
isskel_bool = false;
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recOnWeightedKernel.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/recOnWeightedKernel.m
| 12,345 |
utf_8
|
e93849a49bae2f70b66d852857515f10
|
function [er, optim, opts] = recOnWeightedKernel(varargin)
tic
%% options updates
close all;
if(nargin == 0)
[opts, ~] = initializeOpts;
elseif(nargin == 1)
[opts, ~] = initializeOpts(varargin{1});
elseif(nargin == 2)
[opts, ~] = initializeOpts(varargin{1},varargin{2});
elseif(nargin == 3)
[opts, ~] = initializeOpts(varargin{1},varargin{2},varargin{3});
elseif(nargin == 4)
[opts, ~] = initializeOpts(varargin{1},varargin{2},varargin{3},varargin{4});
elseif(nargin == 5)
[opts, ~] = initializeOpts(varargin{1},varargin{2},varargin{3},varargin{4},varargin{5});
else
disp('error: wrong number of input arguments');
end
%% Options to select
opts.round = 1;
opts.database = 'CMU_amc_regFit';
opts.loadPath = fullfile('../Data/');
opts.loadInPath = fullfile('../Data/');
opts.saveResPath = fullfile('../Data/');
if ~exist(opts.saveResPath,'dir')
mkdir(opts.saveResPath);
end
%% Loading input anf GT
[motIn,~, opts] = loadnPrepareQMot(opts);
load(fullfile([opts.loadPath 'motGT_' opts.subject '_' opts.actName ])); %gt
estP2D = cell2mat(motIn.jointTrajectories2DF);
[cm.fc, cm.cc, cm.alpha_c, cm.kc, cm.Rc_ext, cm.omc_ext, cm.Tc_ext] = readSpicaCalib(opts.pathCalFile);
%% Loading nn
%shape = {'upper';'lower';'left';'right';'pose'};
try
shape = motIn.bodyTypeOrder;
catch
disp('Please load right input');
end
for h = 1:length(shape)
loadPath = opts.loadPath;
idName = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' shape{h} '_obj.mat'];
load([loadPath idName]);
objAll{h,1}.obj = obj;
objAll{h,1}.knn = opts.knn;
objAll{h,1}.knnJoints = opts.allJoints;
objAll{h,1}.bodyType = shape{h};
idNameCM = [opts.subject '_' opts.actName '_' num2str(opts.knn) '_' shape{h} '_camMtx.mat'];
fileCM = [loadPath idNameCM];
load(fileCM);
camMtxAll{h,1} = camMtx;
loadPath = opts.loadPath;
opts.cJno = getShapeJointNum(shape{h},opts.inputDB);
[opts.allJoints, opts.cJoints] = getJointsHE (opts.inputDB,opts.database,opts.cJno,shape{h});
opts = getJointsIdx(opts);
jnts.cJidx{h,1} = opts.cJidx;
jnts.cJidx2{h,1} = opts.cJidx2;
jnts.cJidx3{h,1} = opts.cJidx3;
end
%% reconstruction options
erJoints = 14;
selknn = 256;
nknn = 64; % 64
optim.nknn = nknn;
optim.nrOfSVPosePriorP = 18; %18 % 25 %30
optim.nrOfPrinComps = 18; %18 % 25 %30
optim.w_pose = .35; %20 .60
optim.w_control = .55; %10 .45
optim.w_pose2D = .25; %10 0
optim.w_limb = .065; %10 .05
fprintf('%3d ', optim.w_pose );
fprintf('%3d ', optim.w_control);
fprintf('%3d ', optim.w_pose2D );
fprintf('%3d ', optim.w_limb );
disp(' . . . done');
optim.expPPrior = 0.5;
optim.expMPrior = 0.5;
optim.expSmooth = 0.5;
optim.expControl = 0.5;
optim.jointWeights = 1;
optim.jointWeights2D = 1;
% optim.jointWeights = ones(3*length(opts.allJoints),selknn);
% optim.jointWeights2D = ones(2*length(opts.allJoints),selknn);
optLSQ = optimset( 'Display','none',... %'iter',
'MaxFunEvals',100000,...
'MaxIter',1000,...
'TolFun',0.001,...
'TolX',0.001,...
'LargeScale','on',...
'Algorithm','levenberg-marquardt');
lb = [];
ub = [];
%--------------------------------------------------------------------------
lbb = -0.5;
ubb = 1.5;
startValueMtx = [0;0;0;0;0;0];
windowsize = 4; % for 5 we have to select 4
optimOpts = optimset(...
'Display','off',... % 'Display','iter'
'MaxIter',10000,...
'MaxFunEvals',10000,... % 'PlotFcns', @optimplotx, ...
'Algorithm' , 'trust-region-reflective' , ...%'trust-region-reflective',... %'levenberg-marquardt' , ...
'TolX',0.001);
%% Reconstruction
optim.limlenIdx = H_calLimbLengthIdx(opts.allJoints);
sknnObj = obj;
obj.data = [];
sknnObj.knn = nknn;
sknnObj.data = nan(nknn*3*motIn.njoints, motIn.nframes);
toc
tic
fprintf('Reconstruction: Frame no. ... ');
for f = 1:motIn.nframes
fprintf('%3d ', f);
sObj = objAll{motIn.bodyType(f),1};
sknn = sObj.obj.data(:,f);
sknn = reshape(sknn,3*length(opts.allJoints),[]);
sknn = sknn(:,1:selknn);
for j = 1:length(optim.limlenIdx)
optim.limlen(j,:) = sqrt(sum((sknn(optim.limlenIdx(j,1:3),:) - sknn(optim.limlenIdx(j,4:6),:)).^2));
%optim.limlen(j,:) = sum(sqrt((sknn(optim.limlenIdx(j,1:3),:) - sknn(optim.limlenIdx(j,4:6),:)).^2));
end
%====================================================================== weights regularization
w = motIn.knnWts(1:selknn,f)';
ind = find(w>1);
w(ind) = 0;
w = regulateWeights(w);
[~, wtidx] = sort(w,'descend');
w = w(wtidx(1:nknn));
sw = sum(w);
optim.limlen = optim.limlen(:,wtidx(1:nknn));
sknn = sknn(:,wtidx(1:nknn));
sknnObj.data(:,f) = sknn(:);
%======================================================================
opts.bodyType = sObj.bodyType;
opts.cJno = getShapeJointNum(opts.bodyType,opts.inputDB);
[opts.allJoints, opts.cJoints] = getJointsHE (opts.inputDB,opts.database,opts.cJno,opts.bodyType);
opts = getJointsIdx(opts);
optim.estP2D = estP2D(:,f);
optim.camMtx = cell2mat(camMtxAll{motIn.bodyType(f),1}(f,1));
optim.cJidx2 = opts.cJidx2;
optim.wtPose = w;
%======================================================================
% knn repeats according to the weights for weighted pca
wlm = ceil(w*10);
for k = 1:length(wlm)
nn = repmat(sknn(:,k),1,wlm(k));
nn2d = repmat(sknn(:,k),1,wlm(k));
wt = repmat(w(k),1,wlm(k));
if (k == 1)
nnrep = nn;
nnrep2d = nn2d;
wrep = wt;
else
nnrep = [nnrep,nn];
nnrep2d = [nnrep2d,nn2d];
wrep = [wrep,wt];
end
end
optim.wtPoserep = wrep;
optim.localModelPosePrior_Pos = nnrep'; %sknn';
[optim.coeffs_Pos,score_Pos] = princomp2(optim.localModelPosePrior_Pos);
optim.meanVec_Pos = mean(optim.localModelPosePrior_Pos)';%rec(:,f);%
covMatPosePrior_Pos = computeCovMat(optim.localModelPosePrior_Pos,optim.nrOfSVPosePriorP);
optim.invCovMatPosePrior_Pos = inv(covMatPosePrior_Pos);
startValue = score_Pos(1,1:optim.nrOfPrinComps);
% lb = min([score_Pos(:,1:optim.nrOfPrinComps)*lbb; score_Pos(:,1:optim.nrOfPrinComps)*ubb]);
% ub = max([score_Pos(:,1:optim.nrOfPrinComps)*lbb; score_Pos(:,1:optim.nrOfPrinComps)*ubb]);
Xo = lsqnonlin(@(x) objfunLocal(x,optim,cm),startValue,lb,ub,optLSQ);
recOpt(:,f) = optim.coeffs_Pos(:,1:optim.nrOfPrinComps) * Xo' + optim.meanVec_Pos;
wtidx = [];
w = [];
sknn = [];
optim.limlen = [];
fprintf('\b\b\b\b');
end
fprintf('\b done \n');
toc
recOpt = H_rigidTransformKnn(recOpt,motGT);
motrecOpt = H_mat2cellMot(recOpt,motGT);
[er.er3DPoseOpt, er.er3DPoseOpt.std] = H_cal3DError(motrecOpt,motGT,erJoints);
disp('3D pose error per every five frames : ');
disp(er.er3DPoseOpt.errFr1_5(1,1));
end
function f = objfunLocal(x,optim,cm)
%% local Objective functions
pos_curr = optim.coeffs_Pos(:,1:optim.nrOfPrinComps) * x' + optim.meanVec_Pos;
%% Prior pose term
e_pose = computePrior_local(optim.localModelPosePrior_Pos',pos_curr,optim.wtPoserep,optim.jointWeights,3,optim.expMPrior);
e_pose = e_pose/sqrt(numel(e_pose));
%% control term
est3d = H_regularize(pos_curr);
est3d = optim.camMtx * est3d;
est2d = project_points2Short(est3d,cm.omc_ext,cm.Tc_ext,cm.fc,cm.cc,cm.kc,cm.alpha_c);
est2d = reshape(est2d,[],1);
e_control = optim.estP2D(optim.cJidx2) - est2d(optim.cJidx2);
e_control = reshape(e_control,2,numel(e_control)/2);
e_control = sqrt(sum(e_control.^2));
e_control = reshape(e_control,[],1);
e_control = e_control /sqrt( numel(e_control)); %
for k = 1:length(optim.limlenIdx)
limlen(k,1) = sqrt(sum((pos_curr(optim.limlenIdx(k,1:3)) - pos_curr(optim.limlenIdx(k,4:6))).^2));
end
e_limb = optim.limlen - limlen(:,ones(1,numel(optim.wtPose)));
e_limb = reshape(e_limb,[],1);
e_limb = sqrt(e_limb.^2);
e_limb = e_limb/sqrt(numel(e_limb));
%% Function return vales --------------------------------------------------
f = [optim.w_pose * e_pose;...
optim.w_control * e_control;...
optim.w_limb * e_limb...
];
end
function est3d = H_regularize(est3d)
%%
est3d = reshape(est3d,3,[]);
est3d ([1 3],:) = flipud(est3d([1 3],:)); % '321' ---> '123'
est3d ([2 3],:) = flipud(est3d([2 3],:)); % '123' ---> '132'
est3d(4,:) = 1;
end
function e = computePrior_local(data,query,poseWeights,jointWeights,dofs,exponent)
%%
% Note: Both poseWeights and jointWeights are already normalized!!
diff = data - query(:,ones(1,numel(poseWeights)));
diff = reshape(diff,dofs,numel(diff)/dofs);
diff = sqrt(sum(diff.^2));
diff = reshape(diff,size(data,1)/dofs,numel(poseWeights));
%diff = reshape(diff,size(data,1),numel(poseWeights));
e = jointWeights .* diff;
%e = e.^exponent;
poseWeights = repmat(poseWeights,size(diff,1),[]);
e = e.* poseWeights;
e = reshape(e,[],1);
end
function covMat = computeCovMat(localModel,nrOfSVPosePrior)
%% Compute Covariance Matrix
[U,S,V] = svd(localModel,'econ');
singVal = diag(S);
totalNrOfSV = numel(singVal);
sigma = sum(singVal(nrOfSVPosePrior+1:end).^2)/(totalNrOfSV-nrOfSVPosePrior);
S2 = diag([singVal(1:nrOfSVPosePrior)' sqrt(sigma)*ones(1,(totalNrOfSV-nrOfSVPosePrior))]);
covMat = V*S2*V'/(totalNrOfSV-1);
end
%==========================================================================
function [er, stdev] = H_cal3DError(motrec,motGT,erJoints)
%% error with input
er.errJnts = nan(length(erJoints),motrec.nframes);
er.errFr = nan(motrec.nframes,1);
er.errFrAll = 0;
inMot = cell2mat(motrec.jointTrajectories);
mocapGT = cell2mat(motGT.jointTrajectories);
[er.errJnts, er.errFr, er.errFrAll] = errorHumanEva(inMot,mocapGT,erJoints);
stdev = std(er.errFr,1);
er.errJnts = squeeze(er.errJnts);
er.errFr1_5(1,1) = mean(er.errFr(1:5:end));
er.errFr1_5(2,1) = mean(er.errFr(2:5:end));
er.errFr1_5(3,1) = mean(er.errFr(3:5:end));
er.errFr1_5(4,1) = mean(er.errFr(4:5:end));
er.errFr1_5(5,1) = mean(er.errFr(5:5:end));
er.errFr1_5(1,2) = std(er.errFr(1:5:end),1); % standard ddeviation
er.errFr1_5(2,2) = std(er.errFr(2:5:end),1);
er.errFr1_5(3,2) = std(er.errFr(3:5:end),1);
er.errFr1_5(4,2) = std(er.errFr(4:5:end),1);
er.errFr1_5(5,2) = std(er.errFr(5:5:end),1);
end
function weights = regulateWeights(w)
weights = w-min(w);
%weights = weights/max(weights);
weights = weights/(max(w) - min(w));
end
function limlenIdx = H_calLimbLengthIdx(allJoints)
%% Limb Lengths ids
limbs1 = {'Left Ankle'; 'Left Knee'; 'Left Hip'; 'Left Shoulder'; 'Left Elbow';
'Right Ankle'; 'Right Knee'; 'Right Hip'; 'Right Shoulder'; 'Right Elbow';
'Left Hip'; 'Head'; 'Neck';'Neck'};
limbs2 = {'Left Knee'; 'Left Hip'; 'Left Shoulder'; 'Left Elbow'; 'Left Wrist';
'Right Knee'; 'Right Hip'; 'Right Shoulder'; 'Right Elbow'; 'Right Wrist';
'Right Hip'; 'Neck'; 'Left Shoulder';'Right Shoulder'};
for n = 1: length(limbs1)
idx1 = getSingleJntIdx(limbs1{n},allJoints,3);
idx2 = getSingleJntIdx(limbs2{n},allJoints,3);
limlenIdx(n,:) = [idx1; idx2]';
end
end
function recOr = H_Orient(rec,camMtxAllSel)
for f = 1:size(rec,2)
recTmp = reshape(rec(:,f),3,[]);
recTmp(4,:) = 1;
%[recOr,dYdom,dYdT] = rigid_motion(recTmp,om,T)
tmp = camMtxAllSel{f} * recTmp;
recOr(:,f) = tmp(:);
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recOnWeightedKernelH36Mbm.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/code/recOnWeightedKernelH36Mbm.m
| 12,075 |
utf_8
|
ffb099efa7b685200da7d38736999ad3
|
function [er, optim, opts, motrecOpt] = recOnWeightedKernelH36Mbm(opts)
%% options updates
close all;
% opts_bak = opts;
% if(nargin == 0)
% [opts, skel] = initializeOpts;
% elseif(nargin == 1)
% [opts, skel] = initializeOpts(varargin{1});
% elseif(nargin == 2)
% [opts, skel] = initializeOpts(varargin{1},varargin{2});
% elseif(nargin == 3)
% [opts, skel] = initializeOpts(varargin{1},varargin{2},varargin{3});
% elseif(nargin == 4)
% [opts, skel] = initializeOpts(varargin{1},varargin{2},varargin{3},varargin{4});
% elseif(nargin == 5)
% [opts, skel] = initializeOpts(varargin{1},varargin{2},varargin{3},varargin{4},varargin{5});
% else
% disp('H_error: H_retMain... wring number of input arguments');
% end
% round = 1;
% opts.round = round;
%
% opts.database = 'Human36Mbm'; %'Human36Mbm'\'CMU_H36M';
% opts.exp = 'nogt'; % nogt/gt2D/gt3D
% opts.loadPath = fullfile('../Data/');
% opts.loadIdxPath = fullfile('../Data/');
% opts.loadInPath = fullfile('../Data/');
% opts.saveResPath = fullfile('../Data/');
if ~exist(opts.saveResPath,'dir')
mkdir(opts.saveResPath);
end
%% Loading input anf GT
[motIn, motInN, opts] = loadnPrepareQMot(opts);
load(fullfile([opts.loadPath 'motGT_' opts.subject '_' opts.actName ])); %gt
if(isfield(opts, 'extractFrameNum') == 1)
motGT = extractSingleGT(motGT, opts.extractFrameNum);
end
estP2D = cell2mat(motIn.jointTrajectories2DF);
cm = motIn.camera;
%% Loading nn
%shape = {'upper';'lower';'left';'right';'pose'};
try
shape = motIn.bodyTypeOrder;
catch
disp('Please load right input which has weighted knn ...');
end
for h = 1:length(shape)
loadIdxPath = opts.loadIdxPath;
idName = [opts.subject '_Activity_All_C2_256_' shape{h} '_obj.mat'];
load([loadIdxPath idName]);
objAll{h,1}.obj = obj;
objAll{h,1}.knn = opts.knn;
objAll{h,1}.knnJoints = opts.allJoints;
objAll{h,1}.bodyType = shape{h};
idNameCM = [opts.subject '_Activity_All_C2_256_' shape{h} '_camMtx.mat'];
load([loadIdxPath idNameCM]);
camMtxAll{h,1} = camMtx;
loadIdxPath = opts.loadIdxPath;
opts.cJno = getShapeJointNum(shape{h},opts.inputDB);
[opts.allJoints, opts.cJoints] = getJointsHE (opts.inputDB,opts.database,opts.cJno,shape{h});
opts = getJointsIdx(opts);
jnts.cJidx{h,1} = opts.cJidx;
jnts.cJidx2{h,1} = opts.cJidx2;
jnts.cJidx3{h,1} = opts.cJidx3;
end
%% reconstruction options
erJoints = 14;
selknn = 256;
nknn = 64;
optim.nknn = nknn;
optim.nrOfSVPosePriorP = 18; %25 %30 % number of singular values used for pose term
optim.nrOfPrinComps = 18; %25 %30
optim.w_pose = .35; %20 .60
optim.w_control = .55; %10 .45
optim.w_limb = .065; %10 .05
fprintf('%3d ', optim.w_pose );
fprintf('%3d ', optim.w_control);
fprintf('%3d ', optim.w_limb );
disp(' . . . done');
optim.expPPrior = 0.5;
optim.expMPrior = 0.5;
optim.expSmooth = 0.5;
optim.expControl = 0.5;
optim.jointWeights = 1;
optim.jointWeights2D = 1;
optLSQ = optimset( 'Display','none',... %'iter',
'MaxFunEvals',100000,...
'MaxIter',1000,...
'TolFun',0.001,...
'TolX',0.001,...
'LargeScale','on',...
'Algorithm','levenberg-marquardt');
lb = [];
ub = [];
%% Reconstruction
bt = 'all'; % all/pose
optim.limlenIdx = H_calLimbLengthIdx(opts.allJoints);
tic
fprintf('Reconstruction: Frame no. ... ');
for f = 1:motIn.nframes
fprintf('%3d ', f);
if(strcmp(bt,'all'))
sObj = objAll{motIn.bodyType(f),1};
else
sObj = objAll{1,1};
end
sknn = sObj.obj.data(:,f);
sknn = reshape(sknn,3*length(opts.allJoints),[]);
sknn = sknn(:,1:selknn);
for j = 1:length(optim.limlenIdx)
optim.limlen(j,:) = sqrt(sum((sknn(optim.limlenIdx(j,1:3),:) - sknn(optim.limlenIdx(j,4:6),:)).^2));
end
w = motIn.knnWts(1:selknn,f)';
ind = find(w>1);
w(ind) = 0;
w = regulateWeights(w);
[~, wtidx] = sort(w,'descend');
w = w(wtidx(1:nknn));
optim.limlen = optim.limlen(:,wtidx(1:nknn));
sknn = sknn(:,wtidx(1:nknn));
opts.bodyType = sObj.bodyType;
opts.cJno = getShapeJointNum(opts.bodyType,opts.inputDB);
[opts.allJoints, opts.cJoints] = getJointsHE (opts.inputDB,opts.database,opts.cJno,opts.bodyType);
opts = getJointsIdx(opts);
optim.estP2D = estP2D(:,f);
if(strcmp(bt,'all'))
optim.camMtx = cell2mat(camMtxAll{motIn.bodyType(f),1}(f,1));
else
optim.camMtx = cell2mat(camMtxAll{1,1}(f,1));
end
optim.cJidx2 = opts.cJidx2;
optim.wtPose = w;
wlm = ceil(w*10);
for k = 1:length(wlm)
nn = repmat(sknn(:,k),1,wlm(k));
nn2d = repmat(sknn(:,k),1,wlm(k));
wt = repmat(w(k),1,wlm(k));
if (k == 1)
nnrep = nn;
nnrep2d = nn2d;
wrep = wt;
else
nnrep = [nnrep,nn];
nnrep2d = [nnrep2d,nn2d];
wrep = [wrep,wt];
end
end
optim.wtPoserep = wrep;
optim.localModelPosePrior_Pos = nnrep'; %sknn'; %
[optim.coeffs_Pos,score_Pos] = princomp2(optim.localModelPosePrior_Pos);
optim.meanVec_Pos = mean(optim.localModelPosePrior_Pos)';%rec(:,f);%
covMatPosePrior_Pos = computeCovMat(optim.localModelPosePrior_Pos,optim.nrOfSVPosePriorP);
optim.invCovMatPosePrior_Pos = inv(covMatPosePrior_Pos);
if f > 1
startValue = (optim.coeffs_Pos(:,1:optim.nrOfPrinComps) \ (recOpt(:,f - 1) - optim.meanVec_Pos))';
else
startValue = score_Pos(1,1:optim.nrOfPrinComps);
end
startValue = score_Pos(1,1:optim.nrOfPrinComps);
Xo = lsqnonlin(@(x) objfunLocal(x,optim,cm),startValue,lb,ub,optLSQ);
recOpt(:,f) = optim.coeffs_Pos(:,1:optim.nrOfPrinComps) * Xo' + optim.meanVec_Pos;
wtidx = [];
w = [];
sknn = [];
optim.limlen = [];
fprintf('\b\b\b\b');
end
fprintf('\b done \n');
toc
%% Reconstruct with kernal approach
recOpt = H_rigidTransformKnn(recOpt,motGT);
motrecOpt = H_mat2cellMot(recOpt,motGT);
[er.er3DPoseOpt, er.er3DPoseOpt.std] = H_cal3DError(motrecOpt,motGT,erJoints);
disp(er.er3DPoseOpt.errFrAll);
%motPlay3D('Human36Mbm',motrecOpt);
end
function f = objfunLocal(x,optim,cm)
pos_curr = optim.coeffs_Pos(:,1:optim.nrOfPrinComps) * x' + optim.meanVec_Pos;
e_pose = computePrior_local(optim.localModelPosePrior_Pos',pos_curr,optim.wtPoserep,optim.jointWeights,3,optim.expMPrior);
e_pose = e_pose/sqrt(numel(e_pose));
est3d = H_regularize(pos_curr);
[est2d,~] = H_proj(optim.camMtx,est3d,cm);
est2d = reshape(est2d,[],1);
e_control = optim.estP2D(optim.cJidx2) - est2d(optim.cJidx2);
e_control = reshape(e_control,2,numel(e_control)/2);
e_control = sqrt(sum(e_control.^2));
e_control = reshape(e_control,[],1);
e_control = e_control /sqrt( numel(e_control)); %
for k = 1:length(optim.limlenIdx)
limlen(k,1) = sqrt(sum((pos_curr(optim.limlenIdx(k,1:3)) - pos_curr(optim.limlenIdx(k,4:6))).^2));
end
e_limb = optim.limlen - limlen(:,ones(1,numel(optim.wtPose)));
e_limb = reshape(e_limb,[],1);
e_limb = sqrt(e_limb.^2);
e_limb = e_limb/sqrt(numel(e_limb));
f = [optim.w_pose * e_pose;...
optim.w_control * e_control;...
optim.w_limb * e_limb...
];
end
function est3d = H_regularize(est3d)
%%
est3d = reshape(est3d,3,[]);
est3d([2 3],:) = flipud(est3d([2 3],:));
est3d(4,:) = 1;
end
function e = computePrior_local(data,query,poseWeights,jointWeights,dofs,exponent)
%%
% Note: Both poseWeights and jointWeights are already normalized!!
diff = data - query(:,ones(1,numel(poseWeights)));
diff = reshape(diff,dofs,numel(diff)/dofs);
diff = sqrt(sum(diff.^2));
diff = reshape(diff,size(data,1)/dofs,numel(poseWeights));
%diff = reshape(diff,size(data,1),numel(poseWeights));
e = jointWeights .* diff;
%e = e.^exponent;
poseWeights = repmat(poseWeights,size(diff,1),[]);
e = e.* poseWeights;
e = reshape(e,[],1);
end
function covMat = computeCovMat(localModel,nrOfSVPosePrior)
%% Compute Covariance Matrix
[U,S,V] = svd(localModel,'econ');
singVal = diag(S);
totalNrOfSV = numel(singVal);
sigma = sum(singVal(nrOfSVPosePrior+1:end).^2)/(totalNrOfSV-nrOfSVPosePrior);
S2 = diag([singVal(1:nrOfSVPosePrior)' sqrt(sigma)*ones(1,(totalNrOfSV-nrOfSVPosePrior))]);
covMat = V*S2*V'/(totalNrOfSV-1);
end
%==========================================================================
function [er, stdev] = H_cal3DError(motrec,motGT,erJoints)
%% error with input
er.errJnts = nan(length(erJoints),motrec.nframes);
er.errFr = nan(motrec.nframes,1);
er.errFrAll = 0;
inMot = cell2mat(motrec.jointTrajectories);
mocapGT = cell2mat(motGT.jointTrajectories);
[er.errJnts, er.errFr, er.errFrAll] = errorHumanEva(inMot,mocapGT,erJoints);
stdev = std(er.errFr,1); %
er.errJnts = squeeze(er.errJnts);
er.errFr1_5(1,1) = mean(er.errFr(1:5:end));
er.errFr1_5(2,1) = mean(er.errFr(2:5:end));
er.errFr1_5(3,1) = mean(er.errFr(3:5:end));
er.errFr1_5(4,1) = mean(er.errFr(4:5:end));
er.errFr1_5(5,1) = mean(er.errFr(5:5:end));
er.errFr1_5(1,2) = std(er.errFr(1:5:end),1); % standard ddeviation
er.errFr1_5(2,2) = std(er.errFr(2:5:end),1);
er.errFr1_5(3,2) = std(er.errFr(3:5:end),1);
er.errFr1_5(4,2) = std(er.errFr(4:5:end),1);
er.errFr1_5(5,2) = std(er.errFr(5:5:end),1);
end
function weights = regulateWeights(w)
%% Normalize weights between 0 - 1
weights = w-min(w);
%weights = weights/max(weights);
weights = weights/(max(w) - min(w));
end
function limlenIdx =H_calLimbLengthIdx(allJoints)
%% Limb Lengths ids
limbs1 = {'Left Ankle'; 'Left Knee'; 'Left Hip'; 'Left Shoulder'; 'Left Elbow';
'Right Ankle'; 'Right Knee'; 'Right Hip'; 'Right Shoulder'; 'Right Elbow';
'Left Hip'; 'Head'; 'Neck';'Neck'};
limbs2 = {'Left Knee'; 'Left Hip'; 'Left Shoulder'; 'Left Elbow'; 'Left Wrist';
'Right Knee'; 'Right Hip'; 'Right Shoulder'; 'Right Elbow'; 'Right Wrist';
'Right Hip'; 'Neck'; 'Left Shoulder';'Right Shoulder'};
for n = 1: length(limbs1)
idx1 = getSingleJntIdx(limbs1{n},allJoints,3);
idx2 = getSingleJntIdx(limbs2{n},allJoints,3);
limlenIdx(n,:) = [idx1; idx2]';
end
end
function H_motionPlayer(skel,mot1,mot2)
scalingFactor = 10;
for i = 1:mot1.njoints
mot1.jointTrajectories{i,1}(1,:) = mot1.jointTrajectories{i,1}(1,:)/scalingFactor ;
mot1.jointTrajectories{i,1}(2,:) = mot1.jointTrajectories{i,1}(2,:)/scalingFactor ;
mot1.jointTrajectories{i,1}(3,:) = mot1.jointTrajectories{i,1}(3,:)/scalingFactor ;
mot2.jointTrajectories{i,1}(1,:) = mot2.jointTrajectories{i,1}(1,:)/scalingFactor ;
mot2.jointTrajectories{i,1}(2,:) = mot2.jointTrajectories{i,1}(2,:)/scalingFactor ;
mot2.jointTrajectories{i,1}(3,:) = mot2.jointTrajectories{i,1}(3,:)/scalingFactor ;
end
mot1.boundingBox = computeBoundingBox(mot1);
mot2.boundingBox = computeBoundingBox(mot2);
motionplayerPro('skel',skel,'mot',{mot1,mot2});
end
function [est2d,j3d] = H_proj(x,est3d,cm)
R = makehgtform('xrotate',deg2rad(x(1)),'yrotate',deg2rad(x(2)),'zrotate',deg2rad(x(3)));
R = R(1:3,1:3);
T = [x(4); x(5); x(6)];
R = [R T];
j3d = R*est3d;
% T = T';
% j3d = est3d(1:3,:);
[est2d, ~] = ProjectPointRadial(j3d', cm.R, cm.T, cm.f, cm.c, cm.k, cm.p);
est2d = est2d';
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recMotNG.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/ChaiHodgins/recMotNG.m
| 7,550 |
utf_8
|
5c5a145cfee329a3ba212b265ee261a1
|
function res = recMotNG(skel,mot,priorKnowledge)
% setting variables ----------------------------------------------
dataRep = 'euler'; % quat is bad for prior
controlJoints = [4,9,17,18,20,25,27]; % joints for simulated control signal
selectedJoints = [4,9,19,20,26,27];%[3,4,8,9,12:17,19,20,26,27];%[4,9,20,27,3,8,19,26]; % joints for prior term
k = 25; % number of nearest neighbours
nrOfPrinComps = []; % number of principal components
startFrame = 1;
endFrame = 10;%mot.nframes;
% ----------------------------------------------------------------
% preprocessing data ---------------------------------------------
selectedJointsIndices = jointIDsToMatrixIndices(skel,selectedJoints);
posData = priorKnowledge.mat_pos;
nGraph = priorKnowledge.nGraph;
mot.rootTranslation = zeros(3,mot.nframes);
% mot.rotationQuat{1} = [ones(1,mot.nframes);zeros(3,mot.nframes)];
% mot = convert2euler(skel,mot);
% mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
% mot.boundingBox = computeBoundingBox(mot);
mot = fitRootOrientationsFrameWise(skel,mot);
mot = cutMotion(mot,startFrame,endFrame);
skel = addDOFIDsToSkel(skel);
res.origmot = mot;
optimStruct.origmot = mot;
posData = mat2cell(posData,3*ones(1,size(posData,1)/3),size(posData,2));
posData = cell2mat(posData(controlJoints));
switch dataRep
case 'quat'
rotData = priorKnowledge.mat_quat;
optimStruct.selectedJointsIndices = selectedJointsIndices.quats;
rotations = zeros(size(cell2mat(mot.rotationQuat(mot.animated))));
case 'euler'
rotData = priorKnowledge.mat_euler;
optimStruct.selectedJointsIndices = selectedJointsIndices.euler;
rotations = zeros(size(cell2mat(mot.rotationEuler(mot.animated))));
end
controlSignal = cell2mat(mot.jointTrajectories(controlJoints));
optimStruct.skel = skel;
if isempty(nrOfPrinComps)
nrOfPrinComps = size(rotData,1);
end
optimStruct.nrOfPrinComps = nrOfPrinComps;
optimStruct.qlastlast = [];
optimStruct.qlast = [];
optimStruct.controlJoints = controlJoints;
% quats = cell2mat(mot.rotationQuat(mot.animated));
% optimStruct.qlastlast = quats(:,1);
% optimStruct.qlast = quats(:,2);
% ----------------------------------------------------------------
% setting options for optimization -------------------------------
options = optimset( 'Display','iter',...
'MaxFunEvals',100000,...
'MaxIter',100,...
'TolFun',0.001,...
'PlotFcns', @optimplotx);
% 'LargeScale','off',...%);%,...
lb = [];
ub = [];
% ---------------------------------------------------------------
% estimating nearest neighbours of 0th frame --------------------
distances = sum(abs(posData-repmat(controlSignal(:,1),1,size(posData,2))));
[distancesSorted,indicesSorted] = sort(distances);
NNlast = indicesSorted(1:k); % indices of NN of last synthesized pose
% ---------------------------------------------------------------
for i=startFrame:endFrame
counter = i-startFrame+1;
fprintf('\nReconstructing frame %i (%i/%i)...\n',i,counter,endFrame-startFrame+1);
optimStruct.controlSignal = controlSignal(:,counter);
NNcandidates = getNeighboursNGraph(nGraph,NNlast);
fprintf('Number of NN-candidates: %i\n',length(NNcandidates));
rotData_tmp = rotData(:,NNcandidates);
posData_tmp = posData(:,NNcandidates);
indicesSorted = sortByQueryMetric(rotData_tmp,posData_tmp,optimStruct.qlast,optimStruct.qlastlast,optimStruct.controlSignal);
NNlast = NNcandidates(indicesSorted(1:k));
optimStruct.localModel_rot = rotData(:,NNlast)';
% optimStruct.localModel_pos = posData(:,NNlast)';
[optimStruct.coeffs,optimStruct.score] = princomp(optimStruct.localModel_rot);
optimStruct.meanVec = mean(optimStruct.localModel_rot,1)';
optimStruct.Delta = optimStruct.localModel_rot(:,optimStruct.selectedJointsIndices);
optimStruct.priorFactor = inv(cov(optimStruct.Delta));
% -----------------------------------------------------------
% Es gilt (modulo Rechenungenauigkeit):
% optimStruct.localModel_rot == ...
% optimStruct.score(:,1:nrOfPrinComps) * optimStruct.coeffs(:,1:nrOfPrinComps)' ...
% + repmat(optimStruct.meanVec,nrOfPrinComps,1);
% -----------------------------------------------------------
% optimization for frame i ----------------------------------
startValue = (optimStruct.localModel_rot(1,:)-optimStruct.meanVec') / optimStruct.coeffs(:,1:nrOfPrinComps)';
% startValue = ones(size(optimStruct.meanVec'))/length(optimStruct.meanVec);
X = lsqnonlin(@(x) objfunLocal(x,optimStruct),startValue,lb,ub,options);
% -----------------------------------------------------------
if ~isempty(optimStruct.qlast)
optimStruct.qlastlast = optimStruct.qlast;
end
optimStruct.qlast = optimStruct.coeffs(:,1:nrOfPrinComps) * X'...
+ optimStruct.meanVec;
if ~isempty(optimStruct.qlastlast)
optimStruct.smoothSummand = - 2*optimStruct.qlast + optimStruct.qlastlast;
end
rotations(:,counter) = optimStruct.qlast;
end
res.recmot = buildMotFromRotData(rotations,skel);
end
%% local functions
function f = objfunLocal(x,optimStruct)
rotData_curr = optimStruct.coeffs(:,1:optimStruct.nrOfPrinComps) * x'...
+ optimStruct.meanVec;
q_curr = buildMotFromRotData(rotData_curr,optimStruct.skel);
% ------- control term ------------
e_control = cell2mat(q_curr.jointTrajectories(optimStruct.controlJoints))...
- optimStruct.controlSignal;
e_control = e_control / sqrt(numel(e_control));
% e_control = sum(e_control.^2);
% ------- smoothness term ---------
if ~isempty(optimStruct.qlastlast)
e_smooth = rotData_curr + optimStruct.smoothSummand;
else
e_smooth = zeros(size(rotData_curr));
end
e_smooth = e_smooth / sqrt(numel(e_smooth));
% e_smooth = sum(e_smooth.^2);
% ------- prior term ------------
diffVec = (rotData_curr(optimStruct.selectedJointsIndices)-optimStruct.meanVec(optimStruct.selectedJointsIndices));
e_prior = diffVec'...
* optimStruct.priorFactor...
* diffVec;
e_prior = e_prior / numel(diffVec);
% ------- overall error --------
% f = e_prior;
w_control = 1;%0.8;
w_smooth = 1;%0.2;
w_prior = 1;%0.5;
f = [w_control * e_control;...
w_smooth * e_smooth;...
w_prior * e_prior];
% b = [sum((w_control*e_control).^2) sum((w_smooth *e_smooth ).^2)];
% b = [sum((w_control*e_control).^2) sum((w_smooth *e_smooth ).^2) (w_prior*e_prior).^2];
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
sortByQueryMetric.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/ChaiHodgins/sortByQueryMetric.m
| 872 |
utf_8
|
9ff925af44a4bbf7078f3e093ca8e830
|
% Input:
% - rotData: rotation data of all candidates (for Euler angles of size 59 x #candidates)
% - posData: position data of selected joints of all candidates
% - qlast: rotation data of last synthesized pose
% - qlastlast: rotation data of second last synthesized pose
% - controlSignal: current position data of control markers
function [candidatesSorted,distancesSorted] = sortByQueryMetric(rotData,posData,qlast,qlastlast,controlSignal)
nPoints = size(posData,2);
alpha = 0.8;
beta = 0.2;
control_term = sum((posData-repmat(controlSignal,1,nPoints)).^2);
if ~isempty(qlastlast)
smoothness_term = sum((rotData+repmat(-2*qlast+qlastlast,1,nPoints)).^2);
else
smoothness_term = 0;
end
distances = alpha * control_term + beta * smoothness_term;
[distancesSorted,candidatesSorted] = sort(distances);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recMotNG_new.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/ChaiHodgins/recMotNG_new.m
| 8,130 |
utf_8
|
3e337954f724eaeb097671cfcfcbaf66
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% function recMotNG (REConstruct MOTion using Neighbour Graph)
%
% res = recMotNG(skel,mot,data)
%
% input:
% - skel: skeleton struct
% - mot: mot struct
% - data: struct containing fields 'nGraph', 'quat' and 'pos',
% obtained by data = buildNeighbourGraphC;
% output:
% - res: struct containing fields 'origmot' and 'recmot'
%
% author: Jochen Tautges ([email protected])
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function res = recMotNG_new(skel,mot,data)
% options for reconstruction -------------
% optimStruct.joints = [4,9,17,18,21,25,28]; % joints simulating control data
optimStruct.joints = [4,28]; % joints simulating control data
optimStruct.priorJoints = [3,4,5,8,9,10,12:17,19,20,21,26,27,28]; % joints used for pose prior
optimStruct.nrOfPrinComps = 64; % number of principal components used to represent poses
k = 256; % number of nearest neighbours used to build local model
optimStruct.r = 32; % number of singular values used for prior term
startFrame = 1;
endFrame = mot.nframes;
res.w_control = 1;
res.w_smoothness = 1;
res.w_prior = 10;
% ----------------------------------------
optimStruct.w_control = res.w_control;
optimStruct.w_smoothness = res.w_smoothness;
optimStruct.w_prior = res.w_prior;
% options for optimization ---------------
lb = [];
ub = [];
options = optimset( 'Display','iter',...
'MaxFunEvals',100000,...
'MaxIter',50,...
'TolFun',0.01);%,...
% 'LargeScale','off');%,... % use Levenberg Marquard
% 'PlotFcns', @optimplotx);
% ----------------------------------------
res.skel = skel;
mot = cutMotion(mot,startFrame,endFrame);
res.origmot = mot;
mot.rootTranslation = zeros(3,mot.nframes);
mot.rotationQuat{1}(1,:) = 1;
mot.rotationQuat{1}(2:4,:) = 0;
mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
% mot = fitRootOrientationsFrameWise(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
res.recmot = emptyMotion(mot);
quatrec = zeros(size(data.quat,1),mot.nframes);
optimStruct.jointIDX = jointIDsToMatrixIndices(skel,optimStruct.joints);
optimStruct.priorIDX = jointIDsToMatrixIndices(skel,optimStruct.priorJoints);
dofs = getDOFsFromSkel(skel);
optimStruct.skel = skel;
optimStruct.posLast = [];
optimStruct.posLastLast = [];
controlSignal = cell2mat(mot.jointTrajectories(optimStruct.joints));
% estimating nearest neighbours of 0th frame --------------------
distances = sum(abs(data.pos(optimStruct.jointIDX.pos,:)-repmat(controlSignal(:,1),1,size(data.pos,2))));
[distancesSorted,indicesSorted] = sort(distances);
NNlast = indicesSorted(1:k); % indices of NN of last synthesized pose
% ---------------------------------------------------------------
if isempty(optimStruct.nrOfPrinComps)
optimStruct.nrOfPrinComps = size(data.quat,1);
end
% optimStruct.localModel = data.quat';
% [optimStruct.coeffs,score] = princomp2(optimStruct.localModel);
% optimStruct.meanVec = mean(optimStruct.localModel)';
for i=startFrame:endFrame
counter = i-startFrame+1;
fprintf('\nReconstructing frame %i (%i/%i)...\n',i,counter,endFrame-startFrame+1);
optimStruct.controlSignal = controlSignal(:,counter);
NNcandidates = getNeighboursNGraph(data.nGraph,NNlast);
fprintf('Number of NN-candidates: %i\n',length(NNcandidates));
rotData_tmp = data.quat(optimStruct.jointIDX.quats,NNcandidates);
posData_tmp = data.pos(optimStruct.jointIDX.pos,NNcandidates);
indicesSorted = sortByQueryMetric(rotData_tmp,posData_tmp,optimStruct.posLast,optimStruct.posLastLast,optimStruct.controlSignal);
NNlast = NNcandidates(indicesSorted(1:k));
optimStruct.localModel = data.quat(:,NNlast)';
[optimStruct.coeffs,score] = princomp2(optimStruct.localModel);
optimStruct.meanVec = mean(optimStruct.localModel)';
[U,S,V] = svd(optimStruct.localModel(:,optimStruct.priorIDX.quats));
singVal = diag(S);
sigma = sum(singVal(optimStruct.r+1:end).^2)/(numel(singVal)-optimStruct.r);
S2 = diag([singVal(1:optimStruct.r)' sqrt(sigma)*ones(1,(numel(singVal)-optimStruct.r))]);
C = V*S2*V'/(numel(singVal)-1);
optimStruct.meanVecPrior = mean(optimStruct.localModel(:,optimStruct.priorIDX.quats))';
optimStruct.priorFactor = inv(C);
if counter==1
% choose x of nearest neighbour as startvalue
startValue = score(1,1:optimStruct.nrOfPrinComps);
else
% choose x of last reconstructed pose as startvalue
startValue = (optimStruct.coeffs(:,1:optimStruct.nrOfPrinComps) \ (quatrec(:,counter-1) - optimStruct.meanVec))';
end
% optimization -------------------------------------
X = lsqnonlin(@(x) objfunLocal(x,optimStruct),startValue,lb,ub,options);
% --------------------------------------------------
quatrec(:,counter) = optimStruct.coeffs(:,1:optimStruct.nrOfPrinComps) * X' + optimStruct.meanVec;
if ~isempty(optimStruct.posLast)
optimStruct.qLastLast = optimStruct.qLast;
optimStruct.qLast = quatrec(:,counter);
optimStruct.summand = - 2*optimStruct.qLast + optimStruct.qLastLast;
else
optimStruct.qLast = quatrec(:,counter);
end
end
res.recmot.rootTranslation = res.origmot.rootTranslation;
res.recmot.rotationQuat = mat2cell(quatrec,dofs.quat,i);
res.recmot.rotationQuat{1} = res.origmot.rotationQuat{1};
res.recmot.jointTrajectories = C_forwardKinematicsQuat(skel,res.recmot);
res.recmot.boundingBox = computeBoundingBox(res.recmot);
res.recmot = convert2euler(skel,res.recmot);
res.dist = compareMotions08(skel,res.origmot,res.recmot);
end
%% local functions
function f = objfunLocal(x,optimStruct)
quat_curr = optimStruct.coeffs(:,1:optimStruct.nrOfPrinComps) * x' + optimStruct.meanVec;
mot_curr = buildMotFromRotData(quat_curr,optimStruct.skel);
pos_curr = cell2mat(mot_curr.jointTrajectories);
% control term --------------------------------------------------------
e_control = optimStruct.controlSignal - pos_curr(optimStruct.jointIDX.pos);
e_control = e_control / sqrt(numel(e_control));
% ---------------------------------------------------------------------
% smoothness term -----------------------------------------------------
if ~isempty(optimStruct.posLastLast)
e_smoothness = pos_curr + optimStruct.summand;
else
e_smoothness = zeros(size(pos_curr));
end
e_smoothness = e_smoothness / sqrt(numel(e_smoothness));
% ---------------------------------------------------------------------
% prior term ----------------------------------------------------------
diffVec = quat_curr(optimStruct.priorIDX.quats) - optimStruct.meanVecPrior;
e_prior = diffVec'...
* optimStruct.priorFactor...
* diffVec;
e_prior = e_prior / numel(diffVec);
% ---------------------------------------------------------------------
f = [optimStruct.w_control * e_control;...
optimStruct.w_smoothness * e_smoothness;...
optimStruct.w_prior * e_prior];
% b = [sum((w_control * e_control).^2),...
% sum((w_smoothness * e_smoothness).^2),...
% sum((w_prior * e_prior).^2)];
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
readKinectData.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/kinectRecordings/readKinectData.m
| 2,220 |
utf_8
|
0d752a6e31d824a416f68c48b6eae6b6
|
function [skels,mots] = readKinectData(file)
rawdata = importdata(file);
if ~isempty(rawdata)
timestamps = rawdata(:,1);
skelids = rawdata(:,2);
skeldata = rawdata(:,3:end);
diffskelids = unique(skelids);
numskels = numel(diffskelids);
mots = cell(numskels,1);
skels = cell(numskels,1);
nframes = zeros(numskels,1);
for cs = 1:numskels
mots{cs} = emptyMotion();
skels{cs} = emptySkeleton();
mots{cs}.nframes = sum(skelids==diffskelids(cs));
mots{cs}.filename = [file '_skel_' num2str(diffskelids(cs))];
mots{cs}.timestamps = timestamps(skelids==diffskelids(cs));
mots{cs}.jointTrajectories = skeldata(skelids==diffskelids(cs),:)'*100;
mots{cs}.jointTrajectories = mat2cell(mots{cs}.jointTrajectories, ...
3*ones(20,1),mots{cs}.nframes);
[skels{cs},mots{cs}] = setKinectDefaultValues(skels{cs},mots{cs});
end
else
skels=[];
mots=[];
end
end
function [skel,mot] = setKinectDefaultValues(skel,mot)
% Hardcoded properties of kinect skeleton ...
mot.njoints = 20;
mot.jointNames = {'HIP_CENTER', ...
'SPINE', ...
'SHOULDER_CENTER', ...
'HEAD', ...
'SHOULDER_LEFT', ...
'ELBOW_LEFT', ...
'WRIST_LEFT', ...
'HAND_LEFT', ...
'SHOULDER_RIGHT', ...
'ELBOW_RIGHT', ...
'WRIST_RIGHT', ...
'HAND_RIGHT', ...
'HIP_LEFT', ...
'KNEE_LEFT', ...
'ANKLE_LEFT',...
'FOOT_LEFT',...
'HIP_RIGHT',...
'KNEE_RIGHT',...
'ANKLE_RIGHT',...
'FOOT_RIGHT'}';
mot.boneNames = mot.jointNames;
mot.animated = 1:20;
skel.njoints = mot.njoints;
skel.animated = mot.animated;
skel.boneNames = mot.boneNames;
skel.filename = mot.filename;
skel.nameMap = mot.boneNames;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildTensorActRepStyle.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/buildTensorActRepStyle.m
| 7,136 |
utf_8
|
d6e3ee0b48a4b327d2d03755869165c9
|
function [Tensor]=buildTensorActRepStyle(p,varargin)
% Creates a motion tensor from a given Directories of our
% MocapDB. Each directory corresponds to one style.
% All motions in a given directories are warped and put into the tensor.
% The reference motion is allways the first motion in the first dir.
% author: Bjoern Krueger ([email protected])
% R:\HDM05_3style_example\cut_amc
% If there is a maximum of Reps definied us it, otherwise use all
% motions. ! Can result in a lot of NaN's.
switch nargin
case 2
maxRep =varargin{1};
dataRep='Quat';
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 3
maxRep =varargin{1};
dataRep=varargin{2};
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 4
maxRep =varargin{1};
dataRep=varargin{2};
Styles =varargin{3};
otherwise
error('Wrong number of Args');
end
% Define size of Representation
switch dataRep
case 'Quat'
dimDataRep=4;
case 'Position'
dimDataRep=3;
case 'ExpMap'
dimDataRep=3;
otherwise
error('buildTensorActRepStyle: Wrong Date specified in var: dataRep');
end
% Check if Backslash is included in path and append extension.
dim=size(p);
if(p(dim(2))~='\')
p=[p '\'];
end
ext='*.amc';
% Check if there is a Directory for every Style:
numStyles=size(Styles,2);
for s=1:numStyles
if(~exist([p Styles{1,s}],'dir'))
error(['buildTensorActRepStyle: ' ...
'Dir for Style ' ...
Styles{1,s} ' does not exist!']);
end
end
% Get Lists of files:
for s=1:numStyles
listofFiles{s}=dir([p Styles{1,s} '\' ext]);
numMotions{s} =size(listofFiles{s},1);
end
dimStyleMode =numStyles;
dimActorsMode=5;
dimRepMode =maxRep;
%% Collect Information about the given
%% Files/Styles/Classes
% We need the number of actors, minimum of repetitions
% This are information about the dimension of the
% resulting tensor.
% Known Actors: bd,bk,mm,dg,tr
actors{1}='HDM_bd';
actors{2}='HDM_bk';
actors{3}='HDM_dg';
actors{4}='HDM_mm';
actors{5}='HDM_tr';
numActors=size(actors,2);
reps=zeros(numActors,numStyles);
for s=1:numStyles
%Count repetitions of each actor
for a=1:numActors
reps(a,s)=countActor(listofFiles{s},actors{a});
end
end
% Motions to fit by DTW
skelfile=fullfile(p, Styles{1,1}, [actors{1} '.asf']);
motfile =fullfile(p, Styles{1,1}, listofFiles{1}(1).name);
[fitskel,fitmot]=readMocap(skelfile,motfile);
fprintf('fitmot: ');
fprintf([listofFiles{s}(1).name '\n'])
fitmot=reduceFrameRate(fitskel,fitmot);
Tensor.DataRep=dataRep;
Tensor.numNaturalModes = 3;
Tensor.dimNaturalModes = [dimStyleMode dimActorsMode dimRepMode];
Tensor.numTechnicalModes= 3;
Tensor.dimTechnicalModes= [dimDataRep fitmot.nframes fitmot.njoints];
Tensor.styles=Styles;
% Allocate memory for Tensor
Tensor.data=NaN([Tensor.dimTechnicalModes Tensor.dimNaturalModes]);
for s=1:numStyles
rep=1;
actor=1;
fprintf(' ');
file=1;
% Run throgh list of files
while (file<numMotions{s})
% for file=1:dim(1)
if(actor<=size(actors,2))
if(rep<=reps(actor,s)&&rep<=maxRep)
%Load motion
fprintf('\b\b\b\bRead ');
skelfile=fullfile(p, Styles{1,s}, [actors{actor} '.asf']);
motfile =fullfile(p, Styles{1,s}, listofFiles{s}(file).name);
fprintf(listofFiles{s}(file).name);
[skel,mot]=readMocap(skelfile,motfile);
% Reduce frame rate
mot=reduceFrameRate(skel,mot);
% fit motion
mot=fitMotion(skel,mot);
% Timewarp motion
for c=1:size(listofFiles{s}(file).name,2)
fprintf('\b');
end
fprintf('\b\b\b\b\bWarp');
[mot]=SimpleDTW(fitmot,skel,mot);
% Fill warped motion into tensor
Tensor.rootdata(:,:,s,actor,rep)=mot.rootTranslation;
Tensor.motions{s,actor,rep}=motfile;
Tensor.skeletons{s,actor,rep}=skelfile;
for joint=1:mot.njoints
Tensor.joints{joint,s,actor,rep}=mot.jointNames{joint};
switch dataRep
case 'Quat'
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,s,actor,rep)=mot.rotationQuat{joint};
else
Tensor.data(1,:,joint,s,actor,rep) =ones (1,mot.nframes);
Tensor.data(2:4,:,joint,s,actor,rep)=zeros(3,mot.nframes);
end
case 'Position'
if(~isempty(mot.jointTrajectories{joint}))
Tensor.data(:,:,joint,s,actor,rep)=mot.jointTrajectories{joint};
else
Tensor.data(1:3,:,joint,s,actor,rep)=zeros(3,mot.nframes);
end
case 'ExpMap'
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,s,actor,rep)=quatlog(mot.rotationQuat{joint});
else
Tensor.data(1:3,:,joint,s,actor,rep)=zeros(3,mot.nframes);
end
otherwise
error('buildTensorActRepStyle: Wrong Type specified in var: dataRep\n');
end
end
rep=rep+1;
file=file+1;
else
actor=actor+1;
rep=1;
% file=file-1;
end
else
file=file+1;
end
end
fprintf('\b\b\b\b');
end
Tensor=HOSVDv2(Tensor);
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficientsColumn.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficientsColumn.m
| 3,362 |
utf_8
|
a033daf8edd44b1c910f685662a2d570
|
% FUNCTION findCoefficients searches for optimal coefficients to
% reconstruct a given motion out of a given tensor. It uses the Matlab
% Optimization Toolbox.
% INPUT:
% Tensor: struct: containing data, core, matrices.
% newMot: struct: the motion that should be reconstructed.
% varargin{1}: string: Type of data Representation 'Quat' or 'Position'
% varargin{2}: cell of arrays: Start values for optimization values
% have to correspond to the number of natural and
% technical modes.
%
% OUTPUT:
% X: cell array: coefficients found for best solution
% Y: cell array: X normalized to length 1
% mot: struct: motion reconstructed with X
% d: float: distance, correponding to the used distance
% measure between mot and newMot.
function [X] = findCoefficientsColumn(Tensor,newMot,varargin)
% Align first new Motion like all others!
% [skel,fitmot]=reconstructMotionT(Tensor,[1 1 1]);
% skel = readASF(Tensor.skeletons{1,1});
% newMot=fitMotion(skel,newMot);
% Timewarp motion
%[newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=Tensor.numTechnicalModes;
nrOfNaturalModes =Tensor.numNaturalModes;
% Compute mode-n-product of core tensor and all matrices related to
% technical modes
core_tmp=Tensor.core;
for i=1:nrOfTechnicalModes
core_tmp=modeNproduct(core_tmp,Tensor.factors{i},i);
end
root_tmp=Tensor.rootcore;
for i=1:nrOfTechnicalModes-1
root_tmp=modeNproduct(root_tmp,Tensor.rootfactors{i},i);
end
iter=500;
% Set options for optimization
options = optimset('Display','iter','MaxFunEvals',iter*12,'MaxIter',iter,'TolFun',1e-1);
% Set lower and upper bounds for optimization variable x
dimvec=size(core_tmp);
n=sum(dimvec(nrOfTechnicalModes+1:end));
% lb=-0.5*ones(1,n);
% ub= 1.5*ones(1,n);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
x0=0;
for i=1:nrOfNaturalModes
for j=1:dimvec(nrOfTechnicalModes+i)
x0=[x0 1/dimvec(nrOfTechnicalModes+i)];
end
end
x0=x0(2:end);
setx0=false;
readSkel=true;
switch nargin
case 2
DataRep='Quat';
case 3
DataRep=varargin{1};
case 4
x0=varargin{2};
setx0=true;
DataRep=varargin{1};
case 5
x0=varargin{2};
setx0=true;
DataRep=varargin{1};
skel=varargin{3};
readSkel=false;
otherwise
disp('Wrong number of arguments');
end
if readSkel
skel = readASF(Tensor.skeletons{1,1});
end
% if(setx0)
% lb= min(x0)*ones(1,n);
% ub= max(x0)*ones(1,n);
% else
lb=-inf(1,n);
ub= inf(1,n);
% end
%
% fprintf('\nlower bound x0 upper bound\n');
% disp([lb' x0' ub']);
tmpTensor=Tensor;
tmpTensor.core=core_tmp;
tmpTensor.rootcore=root_tmp;
[X,RESNORM,RESIDUAL] = ...
lsqnonlin(@(x) ...
objfunCol( x,tmpTensor,newMot, ...
nrOfNaturalModes,nrOfTechnicalModes,...
dimvec,skel,DataRep) ...
,x0,lb,ub,options);
% dim=size(Tensor.core);
% dim= dim(Tensor.numTechnicalModes+1:size(dim,2));
% Y=getRowCoefficients(Tensor,vectorToCellArray(X,dim));
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
modeNproduct.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/modeNproduct.m
| 868 |
utf_8
|
16282f02ba80f7e0494e84e44239a479
|
% modeNproduct
% computes the mode-n-product T x_n M
% i.e. T x_n M replaces every mode-n-vector v of T by the product Mv
% example: result = modeNproduct(tensor,matrix,3);
% with tensor (n1 x n2 x n3 x ... x n) and matrix (m x n3)
% author: Jochen Tautges ([email protected])
function result = modeNproduct(Tensor,Matrix,n)
if size(Matrix)==1
result=Tensor*Matrix;
else
nd=ndims(Tensor);
dim=size(Tensor);
if nd<n
nd=nd+1;
dim=[dim 1];
end
order=n:n+nd-1;
order=order-(order>nd)*nd;
dim2=dim;
dim2(n)=size(Matrix,1);
Tensor = permute(Tensor,order);
dim = dim(order);
dim(1) = size(Matrix,1);
result = Matrix*Tensor(:,:);
result = reshape(result,dim);
result = ipermute(result,order);
result = reshape(result,dim2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficients.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficients.m
| 3,421 |
utf_8
|
7854c4bfa177d799469f1223ef221752
|
% FUNCTION findCoefficients searches for optimal coefficients to
% reconstruct a given motion out of a given tensor. It uses the Matlab
% Optimization Toolbox.
% INPUT:
% Tensor: struct: containing data, core, matrices.
% newMot: struct: the motion that should be reconstructed.
% varargin{1}: string: Type of data Representation 'Quat' or 'Position'
% varargin{2}: cell of arrays: Start values for optimization values
% have to correspond to the number of natural and
% technical modes.
%
% OUTPUT:
% X: cell array: coefficients found for best solution
% Y: cell array: X normalized to length 1
% mot: struct: motion reconstructed with X
% d: float: distance, correponding to the used distance
% measure between mot and newMot.
function [X] = findCoefficients(Tensor,newMot,varargin)
% Align first new Motion like all others!
[skel,fitmot]=reconstructMotionT(Tensor,[1 1]);
skel = readASF(Tensor.skeletons{1,1});
newMot=fitMotion(skel,newMot);
% Timewarp motion
[newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=3;
nrOfNaturalModes=ndims(Tensor.core)-nrOfTechnicalModes;
% Compute mode-n-product of core tensor and all matrices related to
% technical modes
core_tmp=Tensor.core;
for i=1:nrOfTechnicalModes
core_tmp=modeNproduct(core_tmp,Tensor.factors{i},i);
end
root_tmp=Tensor.rootcore;
for i=1:nrOfTechnicalModes-1
root_tmp=modeNproduct(root_tmp,Tensor.rootfactors{i},i);
end
% Set options for optimization
options = optimset('Display','iter','MaxFunEvals',2500,'MaxIter',500);
% Set lower and upper bounds for optimization variable x
dimvec=size(core_tmp);
n=sum(dimvec(nrOfTechnicalModes+1:end));
lb=-0.5*ones(1,n);
ub= 1.5*ones(1,n);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
x0=0;
for i=1:nrOfNaturalModes
for j=1:dimvec(nrOfTechnicalModes+i)
x0=[x0 1/dimvec(nrOfTechnicalModes+i)];
end
end
x0=x0(2:end);
switch nargin
case 2
DataRep='Quat';
case 3
DataRep=varargin{1};
case 4
x0=varargin{2};
DataRep=varargin{1};
otherwise
disp('Wrong number of arguments');
end
fprintf('\nlower bound x0 upper bound\n');
disp([lb' x0' ub']);
tmpTensor=Tensor;
tmpTensor.core=core_tmp;
tmpTensor.rootcore=root_tmp;
[Y,RESNORM,RESIDUAL] = ...
lsqnonlin(@(x) ...
objfun( x,tmpTensor,newMot, ...
nrOfNaturalModes,nrOfTechnicalModes,...
dimvec,skel,DataRep) ...
,x0,lb,ub,options);
% Show computed coefficients
for i=1:nrOfNaturalModes
X{i}=Y(1:dimvec(i+nrOfTechnicalModes));
% X2{i}=round(X{i});
Y=Y(dimvec(i+nrOfTechnicalModes)+1:size(Y,2));
end
% Construct motion with computed coefficients and compute mean error
% [skel ,mot] =constructMotion(Tensor,X,skel,DataRep);
% [skel2,mot2]=constructMotion(Tensor,X2);
% d =compareMotions(mot, newMot,DataRep);
% d2=compareMotions(mot2,newMot);
% if d2<d
% X=X2;
% d=d2;
% mot=mot2;
% end
for i=1:nrOfNaturalModes
fprintf('\n X{%i}\n',i);
disp(X{i}');
end
% fprintf('Mean error of joint orientations: E = %.3f degrees.\n',d);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficientsBruteForce.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficientsBruteForce.m
| 5,270 |
utf_8
|
b1834d9e2999fbb6b2eb8182380f1cfb
|
% FUNCTION findCoefficients searches for optimal coefficients to
% reconstruct a given motion out of a given tensor. It uses the Matlab
% Optimization Toolbox.
% INPUT:
% Tensor: struct: containing data, core, matrices.
% newMot: struct: the motion that should be reconstructed.
% varargin{1}: string: Type of data Representation 'Quat' or 'Position'
% varargin{2}: cell of arrays: Start values for optimization values
% have to correspond to the number of natural and
% technical modes.
%
% OUTPUT:
% X: cell array: coefficients found for best solution
% Y: cell array: X normalized to length 1
% mot: struct: motion reconstructed with X
% d: float: distance, correponding to the used distance
% measure between mot and newMot.
function [X,dist] = findCoefficientsBruteForce(Tensor,newMot,varargin)
% Align first new Motion like all others!
[skel,fitmot]=reconstructMotionT(Tensor,[1 1]);
skel = readASF(Tensor.skeletons{1,1});
newMot=fitMotion(skel,newMot);
% Timewarp motion
[newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=3;
nrOfNaturalModes=ndims(Tensor.core)-nrOfTechnicalModes;
% Compute mode-n-product of core tensor and all matrices related to
% technical modes
core_tmp=Tensor.core;
for i=1:nrOfTechnicalModes
core_tmp=modeNproduct(core_tmp,Tensor.factors{i},i);
end
root_tmp=Tensor.rootcore;
for i=1:nrOfTechnicalModes-1
root_tmp=modeNproduct(root_tmp,Tensor.rootfactors{i},i);
end
% Set options for optimization
% options = optimset('Display','iter','MaxFunEvals',10000);
% Set lower and upper bounds for optimization variable x
dimvec=size(core_tmp);
% n=sum(dimvec(nrOfTechnicalModes+1:end));
% lb=-2*ones(1,n);
% ub=2*ones(1,n);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
% x0=0;
% for i=1:nrOfNaturalModes
% for j=1:dimvec(nrOfTechnicalModes+i)
% x0=[x0 1/dimvec(nrOfTechnicalModes+i)];
% end
% end
% x0=x0(2:end);
switch nargin
case 2
DataRep='Quat';
case 3
DataRep=varargin{1};
case 4
x0=varargin{2};
DataRep=varargin{1};
otherwise
disp('Wrong number of arguments');
end
% fprintf('\nlower bound x0 upper bound\n');
% disp([lb' x0' ub']);
tmpTensor=Tensor;
tmpTensor.core=core_tmp;
tmpTensor.rootcore=root_tmp;
X=zeros(1,8);
dist=inf(1,1);
tmp2=inf(1,1);
fprintf('\n');
for c=1:35
fprintf(' ');
end
for x1=1:11
for x2=1:11
% for x3=1:11
% for x4=1:11
% for x5=1:11
for x6=1:11
for x7=1:11
% for x8=1:11
x0(1)=x1/10-0.1;
x0(2)=x2/10-0.1;
x0(3)=0;%x3/10-0.1;
x0(4)=0;%x4/10-0.1;
x0(5)=0;%x5/10-0.1;
x0(6)=x6/10-0.1;
x0(7)=x7/10-0.1;
x0(8)=0;%x8/10-0.1;
for c=1:35
fprintf('\b');
end
fprintf('x0= %1.1f %1.1f %1.1f %1.1f %1.1f %1.1f %1.1f %1.1f', ...
x0(1),x0(2),x0(3),x0(4),x0(5),x0(6),x0(7),x0(8))
tmp= objfun(x0,tmpTensor,newMot, ...
nrOfNaturalModes, ...
nrOfTechnicalModes,...
dimvec,skel,DataRep);
tmp2=sum(tmp(:).*tmp(:));
if(tmp2<dist)
dist=tmp2
X=x0
fprintf('\n');
for c=1:35
fprintf(' ');
end
end
% end
end
end
% end
% end
% end
end
end
% Show computed coefficients
% for i=1:nrOfNaturalModes
% X{i}=Y(1:dimvec(i+nrOfTechnicalModes));
% % X2{i}=round(X{i});
% Y=Y(dimvec(i+nrOfTechnicalModes)+1:size(Y,2));
% end
% Construct motion with computed coefficients and compute mean error
% [skel ,mot] =constructMotion(Tensor,X,skel,DataRep);
% [skel2,mot2]=constructMotion(Tensor,X2);
% d =compareMotions(mot, newMot,DataRep);
% d2=compareMotions(mot2,newMot);
% if d2<d
% X=X2;
% d=d2;
% mot=mot2;
% end
%
% for i=1:nrOfNaturalModes
% Y{i}=X{i}/(sqrt(sum(X{i}.*X{i})));
% fprintf('\n Y{%i}\n',i);
% disp(Y{i}');
% end
% fprintf('Mean error of joint orientations: E = %.3f degrees.\n',d);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficientsColumnRoot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficientsColumnRoot.m
| 3,148 |
utf_8
|
911d39058f655f579d6f39d1d8de486e
|
% FUNCTION findCoefficients searches for optimal coefficients to
% reconstruct a given motion out of a given tensor. It uses the Matlab
% Optimization Toolbox.
% INPUT:
% Tensor: struct: containing data, core, matrices.
% newMot: struct: the motion that should be reconstructed.
% varargin{1}: string: Type of data Representation 'Quat' or 'Position'
% varargin{2}: cell of arrays: Start values for optimization values
% have to correspond to the number of natural and
% technical modes.
%
% OUTPUT:
% X: cell array: coefficients found for best solution
% Y: cell array: X normalized to length 1
% mot: struct: motion reconstructed with X
% d: float: distance, correponding to the used distance
% measure between mot and newMot.
function [X,Y] = findCoefficientsColumnRoot(Tensor,newMot,varargin)
% Align first new Motion like all others!
% [skel,fitmot]=reconstructMotionT(Tensor,[1 1 1]);
skel = readASF(Tensor.skeletons{1,1});
newMot=fitMotion(skel,newMot);
% Timewarp motion
%[newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=Tensor.numTechnicalModes-1;
nrOfNaturalModes =Tensor.numNaturalModes;
for i=1:5%nrOfTechnicalModes-1
Tensor.rootcore=modeNproduct(Tensor.rootcore,Tensor.rootfactors{i},i);
end
iter=20000;
% Set options for optimization
options = optimset('Display','iter','MaxFunEvals',iter*5,'MaxIter',iter,'TolFun',1e-3);
% Set lower and upper bounds for optimization variable x
dimvec=Tensor.dimNaturalModes;
n=sum(dimvec(nrOfTechnicalModes:end));
% lb=-0.5*ones(1,n);
% ub= 1.5*ones(1,n);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
x0=0;
for i=1:nrOfNaturalModes
for j=1:dimvec(i)
x0=[x0 1/dimvec(i)];
end
end
x0=x0(2:end);
% x0=[];
% for i=1:nrOfNaturalModes
% y0=[];
% for j=1:dimvec(i)
% y0=[y0 1/dimvec(i)];
% end
% x0=[x0 (Tensor.factors{nrOfTechnicalModes+i}'*y0')'];
% end
setx0=false;
switch nargin
case 2
DataRep='Quat';
case 3
DataRep=varargin{1};
case 4
x0=varargin{2};
setx0=true;
DataRep=varargin{1};
otherwise
disp('Wrong number of arguments');
end
% if(setx0)
% lb= min(x0)*ones(1,n);
% ub= max(x0)*ones(1,n);
% else
lb=-inf(1,n);
ub= inf(1,n);
% end
%
% fprintf('\nlower bound x0 upper bound\n');
% disp([lb' x0' ub']);
%tmpTensor=Tensor;
%tmpTensor.rootcore=root_tmp;
[X,RESNORM,RESIDUAL] = ...
lsqnonlin(@(x) ...
objfunColRoot( x,Tensor,newMot, ...
nrOfNaturalModes, ...
nrOfTechnicalModes,...
dimvec,skel,DataRep), ...
x0,lb,ub,options);
dim= size(Tensor.rootcore);
dim= dim (Tensor.numTechnicalModes+1:size(dim,2));
Y=getRowCoefficients(Tensor,X);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructMotionCut.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/reconstructMotionCut.m
| 3,317 |
utf_8
|
6be9dfe9d14bb5f985dc1e5f70d1332a
|
function resultX=reconstructMotionCut(skel,mot,tensors)
mot=changeFrameRate(skel,mot,30);
info=filename2info(mot.filename);
styles{1}=info.motionCategory;
% styles{2}=info.motionCategory;
TensorID=findTensorForStyles(styles,tensors);
% for subClassInd=1:tensors{TensorID}.dimNaturalModes(1)
tensors{TensorID}.addSkel=skel;
[fitSkel, fitMot] = extractMotion(tensors{TensorID}, ...
tensors{TensorID}.DTW.refMotID);
[gdm, warpPath, ldm] = pointCloudDTW_pos(fitMot, mot, 2);
segMot=warpMotion(warpPath,skel,mot);
% % % ldm = ldm/max(ldm(:));
% % %
% % % parameter.vis = 0;
% % %
% % % % Anpassen!!!
% % %
% % % parameter.match_thresh = 0.5*fitMot.nframes;
% % %
% % % parameter.match_endExclusionForward = 1;
% % % parameter.match_startExclusionBackward = 1;
% % %
% % % hits = retrieveSubsequenceDTWHits(ldm, parameter);
% fprintf('\n\n\nSubsequence DTW found %i Hits!\n\n\n', size(hits,2));
recCount=0;
% for hitInd=1:size(hits,2)
% plotDTWpathNice(ldm,flipud(hits(1, hitInd).match'));
% drawnow();
% warpPath = convertWarpingPath(flipud(hits(1, hitInd).match'));
% segMot = warpMotion(warpPath,skel,mot);
segMot = fitMotion(skel, segMot);
set = defaultSet_eg08;
set.windowLength = segMot.nframes;
set.warping = 0;
recCount = recCount + 1;
resultX{recCount} = emptyResultStruct();
resultX{recCount}.amc = fullfile(info.amcpath,info.amcname);
resultX{recCount}.asf = info.asfname;
resultX{recCount}.startFrame = 1;%hits(1, hitInd).frame_first_matched;
resultX{recCount}.endFrame = mot.nframes;%hits(1, hitInd).frame_last_matched;
resultX{recCount}.motionClass = styles;
resultX{recCount}.styles = tensors{TensorID}.styles;
resultX{recCount}.orgMot = mot;% warpMotion(convertWarpingPath(fliplr(flipud(hits(1, hitInd).match'))),skel,segMot);
fprintf('\nReconstruction { %i } :\n\n',recCount);
resultX{recCount}.res = recMot_eg08(tensors{TensorID},skel,segMot,set);
resultX{recCount}.distUnWarp = compareMotions_eg08(resultX{recCount}.res.origMot,resultX{recCount}.res.recMot);
resultX{recCount}.recMotUnWarp = warpMotion(fliplr(warpPath),skel,resultX{recCount}.res.recMot);
resultX{recCount}.distUnWarp = compareMotions_eg08(resultX{recCount}.orgMot,resultX{recCount}.recMotUnWarp);
end
% end
% end
function listOfTensors = findTensorForStyles(styles, tensors)
listOfTensors=[];
for tensorID=1:size(tensors,2)
positions=0;
for styleID=1:size(tensors{tensorID}.styles,2)
findres=cell2mat(strfind(styles,tensors{tensorID}.styles{styleID}));
if ~isempty(findres)
positions=positions+findres;
end
end
if positions>0
listOfTensors=[listOfTensors tensorID];
end
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildTensorFromDir.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/buildTensorFromDir.m
| 3,528 |
utf_8
|
904870213c131d96a7c4f53dffa8f61e
|
function [Tensor]=buildTensorFromDir(p)
% Creates a motion tensor from a given Directory of our MocapDB
% All motions in a given directory are warped and put into the tensor.
% The reference motion is allways the first motion in dir.
% author: Bjoern Krueger ([email protected])
%% Check if Backslash is included in path and append extension.
dim=size(p);
if(p(dim(2))~='\')
p=[p '\'];
end
ext='*.amc';
%% Get List of files
files=dir([p ext]);
dim=size(files);
%% Collect Information about the given Files
% We need the number of actors, minimum of repetitions
% This are information about the dimension of the
% resulting tensor.
% Known Actors: bd,bk,mm,dg,tr
actors{1}='HDM_bd';
actors{2}='HDM_bk';
actors{3}='HDM_dg';
actors{4}='HDM_mm';
actors{5}='HDM_tr';
%Count repetitions of each actor
reps(1)=countActor(files,actors{1});
reps(2)=countActor(files,actors{2});
reps(3)=countActor(files,actors{3});
reps(4)=countActor(files,actors{4});
reps(5)=countActor(files,actors{5});
if (min(reps)==0)
[numReps,I]=min(reps);
reps(I)=max(reps);
numReps=min(reps);
else
numReps=min(reps);
end
%% Start reading information
rep=1;
actor=1;
% Motions to fit by DTW
skelfile=[p actors{1} '.asf'];
motfile =[p files(1).name];
[fitskel,fitmot]=readMocap(skelfile,motfile);
fprintf('fitmot: ');
fprintf([files(1).name '\n'])
fitmot=reduceFrameRate(fitskel,fitmot);
Tensor.data=NaN(4,fitmot.nframes,fitmot.njoints,size(actors,2),numReps);
fprintf('Motion ');
% Go through all files
for file=1:dim(1)
fprintf('\b\b\b');
fprintf('%2i ',file);
% Check if motion has to be read
if(actor<=size(actors,2))
correctActor=strfind(files(file).name,actors{actor});
else
correctActor=0;
end
if(isempty(correctActor))
correctActor=0;
end
if ((rep<=numReps)&&correctActor==1)
% read motion
fprintf('Read ');
skelfile=[p actors{actor} '.asf'];
motfile=[p files(file).name];
fprintf(files(file).name);
[skel,mot]=readMocap(skelfile,motfile);
mot=reduceFrameRate(skel,mot);
% Timewarp motion
for c=1:size(files(file).name,2)
fprintf('\b');
end
fprintf('\b\b\b\b\bWarp');
[mot]=SimpleDTW(fitmot,skel,mot);
% Fill warped motion into tensor
Tensor.motions{actor,rep}=motfile;
for joint=1:mot.njoints
Tensor.joints{joint,actor,rep}=mot.jointNames{joint};
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,actor,rep)=mot.rotationQuat{joint};
else
Tensor.data(1,:,joint,actor,rep) =ones (1,mot.nframes);
Tensor.data(2:4,:,joint,actor,rep)=zeros(3,mot.nframes);
end
end
rep=rep+1;
fprintf('\b\b\b\b');
else
actor=actor+1;
rep=1;
end
end
fprintf('\n');
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildTensorFromDir2.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/buildTensorFromDir2.m
| 5,307 |
utf_8
|
9e9cd9ad5ec40658a7c6a9c508f9cc8c
|
function [Tensor]=buildTensorFromDir2(p,varargin)
% Creates a motion tensor from a given Directory of our MocapDB
% All motions in a given directory are warped and put into the tensor.
% The reference motion is allways the first motion in dir.
% author: Bjoern Krueger ([email protected])
%% Check if Backslash is included in path and append extension.
dim=size(p);
if(p(dim(2))~='\')
p=[p '\'];
end
ext='*.amc';
%% Get List of files
files=dir([p ext]);
dim=size(files);
%% Collect Information about the given Files
% We need the number of actors, minimum of repetitions
% This are information about the dimension of the
% resulting tensor.
% Known Actors: bd,bk,mm,dg,tr
actors{1}='HDM_bd';
actors{2}='HDM_bk';
actors{3}='HDM_dg';
actors{4}='HDM_mm';
actors{5}='HDM_tr';
%Count repetitions of each actor
reps(1)=countActor(files,actors{1});
reps(2)=countActor(files,actors{2});
reps(3)=countActor(files,actors{3});
reps(4)=countActor(files,actors{4});
reps(5)=countActor(files,actors{5});
% If there is a maximum of Reps definied us it, otherwise use all
% motions. ! Can result in a lot of NaN's.
switch nargin
case 1
maxRep =max(reps);
dataRep='Quat';
case 2
maxRep=varargin{1};
dataRep='Quat';
case 3
maxRep=varargin{1};
dataRep=varargin{2};
otherwise
error('Wrong number of Args');
end
switch dataRep
case 'Quat'
dimDataRep=4;
case 'Position'
dimDataRep=3;
case 'ExpMap'
dimDataRep=3;
otherwise
error('buildTensorFromDir2: Wrong Date specified in var: dataRep');
end
% Motions to fit by DTW
skelfile=[p actors{1} '.asf'];
motfile =[p files(1).name];
[fitskel,fitmot]=readMocap(skelfile,motfile);
fprintf('fitmot: ');
fprintf([files(1).name '\n'])
fitmot=reduceFrameRate(fitskel,fitmot);
% Allocate memory for Tensor
Tensor.data=NaN(dimDataRep,fitmot.nframes,fitmot.njoints,size(actors,2),maxRep);
rep=1;
actor=1;
fprintf(' ');
file=1;
% Run throgh list of files
while (file<dim(1))
% for file=1:dim(1)
if(actor<=size(actors,2))
if(rep<=reps(actor)&&rep<=maxRep)
%Load motion
fprintf('\b\b\b\bRead ');
skelfile=[p actors{actor} '.asf'];
motfile=[p files(file).name];
fprintf(files(file).name);
[skel,mot]=readMocap(skelfile,motfile);
% Reduce frame rate
mot=reduceFrameRate(skel,mot);
% fit motion
mot=fitMotion(skel,mot);
% Timewarp motion
for c=1:size(files(file).name,2)
fprintf('\b');
end
fprintf('\b\b\b\b\bWarp');
[mot]=SimpleDTW(fitmot,skel,mot);
% Fill warped motion into tensor
Tensor.motions{actor,rep}=motfile;
Tensor.skeletons{actor,rep}=skelfile;
Tensor.rootdata(:,:,actor,rep)=mot.rootTranslation;
for joint=1:mot.njoints
Tensor.joints{joint,actor,rep}=mot.jointNames{joint};
switch dataRep
case 'Quat'
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,actor,rep)=mot.rotationQuat{joint};
else
Tensor.data(1,:,joint,actor,rep) =ones (1,mot.nframes);
Tensor.data(2:4,:,joint,actor,rep)=zeros(3,mot.nframes);
end
case 'Position'
if(~isempty(mot.jointTrajectories{joint}))
Tensor.data(:,:,joint,actor,rep)=mot.jointTrajectories{joint};
else
Tensor.data(1:3,:,joint,actor,rep)=zeros(3,mot.nframes);
end
case 'ExpMap'
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,actor,rep)=quatlog(mot.rotationQuat{joint});
else
Tensor.data(1:3,:,joint,actor,rep)=zeros(3,mot.nframes);
end
otherwise
error('buildTensorFromDir2: Wrong Date specified in var: dataRep');
end
end
rep=rep+1;
file=file+1;
else
actor=actor+1;
rep=1;
% file=file-1;
end
else
file=file+1;
end
end
fprintf('\b\b\b\b');
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructMotionT.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/reconstructMotionT.m
| 1,907 |
utf_8
|
9b107df5e11c2c5d4f447ca3b818680a
|
% function reconstructMotion
% reconstructs/approximates an original motion from a given core tensor of
% arbitrary order and related matrices (obtained by HOSVD)
% recMotion = reconstructMotion(core,factors,rows,varargin)
%
% example: reconstructMotion(Tensor,[2,2,1]) for 3 natural modes
%
% Remark: Use the tucker function from the n-way toolbox to perform HOSVD
% and to obtain core and factors:
% [factors,core] = tucker(Tensor, [n1 n2 n3 ... n_m]),
% with Tensor being of m-th order
% type "help tucker" for more information
function [skel,mot] = reconstructMotionT(Tensor,rows,varargin)
DataRep=Tensor.DataRep;
skel = readASF(Tensor.skeletons{rows(1),rows(2),1});
nrOfDim = size(Tensor.factors,2);
nrOfNaturalModes = size(rows,2);
nrOfTechnicalModes = nrOfDim-nrOfNaturalModes;
nrOfDimRoot = size(Tensor.rootfactors,2);
nrOfTechnicalModesRoot = nrOfDimRoot-nrOfNaturalModes;
for i=1:nrOfTechnicalModes
Tensor.core = modeNproduct(Tensor.core,Tensor.factors{i},i);
end
for i=1:nrOfTechnicalModesRoot
Tensor.rootcore = modeNproduct(Tensor.rootcore,Tensor.rootfactors{i},i);
end
for i=nrOfDim:-1:nrOfTechnicalModes+1
Tensor.data = modeNproduct(Tensor.core,Tensor.factors{i}(rows(i-nrOfTechnicalModes),:),i);
end
for i=nrOfDimRoot:-1:nrOfTechnicalModesRoot+1
Tensor.rootcore = modeNproduct(Tensor.rootcore,Tensor.rootfactors{i}(rows(i-nrOfTechnicalModesRoot),:),i);
end
mot=createMotionFromCoreTensor(Tensor,skel,true,true,DataRep);
% dims=size(Tensor.core);
% mot = emptyMotion;
%
% mot.njoints=dims(3);
% mot.nframes=dims(2);
%
% mot.rootTranslation=Tensor.rootcore(:,:);
%
% for joint=1:mot.njoints
% mot.rotationQuat{1,joint}=Tensor.core(:,:,joint);
% end
%
% mot.samplingRate=30;
% mot.frameTime=1/30;
% mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
% mot.boundingBox = computeBoundingBox(mot);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructMotion6D.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/reconstructMotion6D.m
| 1,499 |
utf_8
|
802c28399e1376e0256b07775247fcf8
|
% function reconstructMotion
% reconstructs/approximates an original motion from a given core tensor of
% arbitrary order and related matrices (obtained by HOSVD)
% recMotion = reconstructMotion(core,factors,rows,varargin)
%
% example: reconstructMotion(Tensor,[2,2,1]) for 3 natural modes
%
% Remark: Use the tucker function from the n-way toolbox to perform HOSVD
% and to obtain core and factors:
% [factors,core] = tucker(Tensor, [n1 n2 n3 ... n_m]),
% with Tensor being of m-th order
% type "help tucker" for more information
function [skel,mot] = reconstructMotion6D(Tensor,rows,varargin)
skel = readASF(Tensor.skeletons{rows(1),rows(2),1});
nrOfDim = Tensor.numNaturalModes+Tensor.numTechnicalModes;
nrOfNaturalModes = Tensor.numNaturalModes;
nrOfTechnicalModes = Tensor.numTechnicalModes;
nrOfDimRoot = size(Tensor.rootfactors,2);
nrOfTechnicalModesRoot = nrOfDimRoot-nrOfNaturalModes;
for i=1:nrOfTechnicalModes
Tensor.core = modeNproduct(Tensor.core,Tensor.factors{i},i);
end
for i=1:nrOfTechnicalModesRoot
Tensor.rootcore = modeNproduct(Tensor.rootcore,Tensor.rootfactors{i},i);
end
for i=nrOfTechnicalModes+1:nrOfDim
Tensor.core = modeNproduct(Tensor.core,Tensor.factors{i}(rows(i-nrOfTechnicalModes),:),i);
end
for i=nrOfTechnicalModes:nrOfDim-1
Tensor.rootcore = modeNproduct(Tensor.rootcore,Tensor.rootfactors{i}(rows(i-nrOfTechnicalModesRoot),:),i);
end
mot=createMotionFromCoreTensor(Tensor,skel,true,true,'ExpMap');
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
compareMotions.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/compareMotions.m
| 9,628 |
utf_8
|
2f1c2b1282b0d03b31ea9c4922351fbd
|
% FUNCTION compareMotions compares to given motions. It returns a matrix of
% distances for joints(rows) and frames(columns).
% INPUT:
% mot: struct: motion
% mot1: struct: motion that shold be compared to mot
% varargin{1}: string: Defines which distance measurement should
% be used.
%
% OUTPUT:
% result: matrix: containig distances for every joint for every
% frame. Most optimization tools from Matlabs
% Optimization Toolbox can use this and make summation
% and calculating squares implicit.
function result=compareMotions(mot,mot1,varargin)
existSkel=false;
plotSteps=true;
switch nargin
case 2
dataRep='Quat';
case 3
dataRep=varargin{1};
case 4
dataRep=varargin{1};
skel =varargin{2};
existSkel=true;
case 5
dataRep=varargin{1};
skel =varargin{2};
existSkel=true;
plotSteps=varargin{3};
otherwise
error('compareMotions: Wrong number of arguments!\n');
end
% result=zeros(mot.nframes,mot.njoints);
switch dataRep
case 'Quat'
for i=2:mot.njoints
% result(:,i)=QuatDistRot(i,mot,mot1);
% result(:,i)=QuatDistAxisJoint(i,mot,mot1);
result(:,i)=QuatDist(i,mot,mot1);
% result(:,i)=QuatDistAbs(i,mot,mot1);
% resultP(:,i)=PosDist(i,mot,mot1);
end
case 'Position'
for i=2:mot.njoints
result(:,i)=PosDist(i,mot,mot1);
end
if plotSteps
plot3(mot1.jointTrajectorie{5}','.');
hold on;
plot3( mot.jointTrajectorie{5}');
hold off;
drawnow();
end
case 'ExpMap'
for i=[3 8 19 26]%[2:4 7:9 12:14 18:21 25:28]
result(:,i)=QuatDistRot(i,mot,mot1);
end
if plotSteps
subplot(2,1,1)
plot(mot1.rotationQuat{4}','.');
hold on;
plot( mot.rotationQuat{4}');
hold off;
axis([0 mot.nframes -0.5 1.2])
subplot(2,1,2)
plot(mot1.rotationQuat{20}','.');
hold on;
plot( mot.rotationQuat{20}');
hold off;
axis([0 mot.nframes -0.5 1.2])
drawnow();
% I = getframe(gcf);
% imwrite(I.cdata, ['c:\opt_vid\opt_' datestr(now, 30) '.png']);
end
case 'jointAngle'
if existSkel
JointAngles1 = dataAcquisition(skel, mot, selectFeatures('jointAngle'),false);
JointAngles2 = dataAcquisition(skel, mot1, selectFeatures('jointAngle'),false);
tmp = abs(JointAngles1-JointAngles2);
if plotSteps
plot(JointAngles2','.');
hold on;
plot( JointAngles1');
hold off;
axis([0 mot.nframes 0 pi])
drawnow();
end
result = tmp.*tmp;
else
error(['compareMotions: if jointAngles are compared a '...
'skeleton should be given!\n']);
end
case 'Acce'
mot =addAccToMot(mot);
mot1=addAccToMot(mot1);
counter=0;
% for j=1:dimTechnicalModes(3)
% for j=[5,10,15,22,29]
for j=1:30
for i=1:size(mot.jointAccelerations{j},2)
counter=counter+1;
m {counter,1}=mot .jointAccelerations{j}(:,i);
mRec{counter,1}=mot1.jointAccelerations{j}(:,i);
end
end
result=distVector_pointCloudDistance(m,mRec,0);
if plotSteps
subplot(2,1,1);
joint=5;
plot( mot.jointAccelerations{joint}');
hold on;
plot( mot1.jointAccelerations{joint}','.');
boxsize=2000;
axis([0 mot.nframes -boxsize boxsize]);
hold off;
subplot(2,1,2);
joint=20;
plot( mot.jointAccelerations{joint}');
hold on;
plot( mot1.jointAccelerations{joint}','.');
boxsize=2000;
axis([0 mot.nframes -boxsize boxsize]);
hold off;
drawnow;
end
otherwise
error('compareMotions: Wrong type of Data specified in var: dataRep\n');
end
end
function res=QuatDistRot(i,mot,mot1)
res=zeros(mot.nframes,1);
if(~isempty(mot.rotationQuat{i})&&~isempty(mot1.rotationQuat{i}))
res=distS3(mot.rotationQuat{i},mot1.rotationQuat{i},0.001);
end
end
function res=QuatDistAxisJoint(i,mot,mot1)
res=zeros(mot.nframes,1);
if(~isempty(mot.rotationQuat{i})&&~isempty(mot1.rotationQuat{i}))
smo=size(mot.rotationQuat{i},2);
sm1=size(mot1.rotationQuat{i},2);
if(sm1<smo)
smo=sm1;
end
rot=acosd(mot.rotationQuat{i}(1,1:smo))-real(acos(mot1.rotationQuat{i}(1,1:smo)));
rotR=abs(real(rot));
% rotI=abs(imag(rot));
% rot=abs(acos(mot.rotationQuat{i}(1,1:smo))*2-acos(mot1.rotationQuat{i}(1,1:smo))*2);
x1=mot.rotationQuat{i}(2:4,1:smo);
x2=mot1.rotationQuat{i}(2:4,1:smo);
for i=1:size(x1,2)
if(norm(x1(:,i))>0)
x1(:,i)=x1(:,i)/norm(x1(:,i));
end
if(norm(x2(:,i))>0)
x2(:,i)=x2(:,i)/norm(x2(:,i));
end
end
angR=(1-dot(x1,x2))*180/pi;
% angI=1-dot(imag(x1),imag(x2));
% ang=1-dot(mot.rotationQuat{i}(2:4,1:smo),mot1.rotationQuat{i}(2:4,1:smo));
res=rotR+angR;%+rotI+angI;
end
end
function res=PosDist(i,mot,mot1)
res=zeros(mot.nframes,1);
if(~isempty(mot.jointTrajectories{i})&&~isempty(mot1.jointTrajectories{i}))
smo=size(mot.jointTrajectories{i},2);
sm1=size(mot1.jointTrajectories{i},2);
if(sm1<smo)
smo=sm1;
end
res=abs(mot.jointTrajectories{i}(1,1:smo)-mot1.jointTrajectories{i}(1,1:smo))';
end
end
function res=QuatDist(i,mot,mot1)
res=zeros(mot.nframes,1);
if(~isempty(mot.rotationQuat{i})&&~isempty(mot1.rotationQuat{i}))
smo=size(mot.rotationQuat{i},2);
sm1=size(mot1.rotationQuat{i},2);
if(sm1<smo)
smo=sm1;
end
res=1-dot(mot.rotationQuat{i}(:,1:smo),mot1.rotationQuat{i}(:,1:smo));
end
end
function res=QuatDistAbs(i,mot,mot1)
res=zeros(mot.nframes,1);
if(~isempty(mot.rotationQuat{i})&&~isempty(mot1.rotationQuat{i}))
smo=size(mot.rotationQuat{i},2);
sm1=size(mot1.rotationQuat{i},2);
if(sm1<smo)
smo=sm1;
end
res=sum(abs(mot.rotationQuat{i}(:,1:smo)-mot1.rotationQuat{i}(:,1:smo)),1);
end
end
function res=QuatDistJochen(mot1,mot2) % = compareMotions_jt
for i=1:mot1.njoints
if (~(isempty(mot1.rotationQuat{i})) && ~(isempty(mot2.rotationQuat{i})))
res(i,1) = mean(real(acosd(dot(mot1.rotationQuat{i},mot2.rotationQuat{i}))*2));
end
end
end
% result=sum(result,2)/mot.njoints;
%fprintf('r= %3.3f\n',result);
%Code-Halde:
% for i=2:mot.njoints
% if(~isempty(mot.rotationQuat{i})&&~isempty(mot1.rotationQuat{i}))
% smo=size(mot.rotationQuat{i},2);
% sm1=size(mot1.rotationQuat{i},2);
% if(sm1<smo)
% smo=sm1;
% end
%
% tmp=dot(mot.rotationQuat{i}(:,1:smo),mot1.rotationQuat{i}(:,1:smo));
% j=i*4-3;
% motion1(:,j:j+3)= mot.rotationQuat{i}(:,1:smo)';
% motion2(:,j:j+3)=mot1.rotationQuat{i}(:,1:smo)';
% % result(i)=sum(tmp)/mot.nframes;%real(acosd(sum(tmp,2)/mot.nframes));
% % result(i)=real(acosd(sum(tmp,2)/mot.nframes));
% end
% end
%
% frames=smo;
% dofs=mot.njoints;
% m=motion1.*motion2;
% colsum=0;
% for i=4:4:size(m,2)-3
% colsum=colsum+real(acosd(sum(m(:,i:i+3)')')*2);
% end
% result=sum(colsum)/(frames*dofs);
% result(i)=sum(tmp)/mot.nframes;
% tmp=dot(mot.rotationQuat{i}(:,1:smo),mot1.rotationQuat{i}(:,1:smo));
% tmp=(ones(size(tmp))-tmp);
% % a=sqrt(sum(mot.rotationQuat{i}(:,1:smo).*mot.rotationQuat{i}(:,1:smo)));
% % b=sqrt(sum(mot1.rotationQuat{i}(:,1:smo).*mot1.rotationQuat{i}(:,1:smo)));
% % c=a.*b;
% % tmp=tmp./c;
% % tmp=real(acosd(tmp));
% result(i)=sum(tmp)/mot.nframes;%real(acosd(sum(tmp,2)/mot.nframes));
% %
% tmp=abs(mot.rootTranslation-mot1.rootTranslation);
% result=sum(tmp,2);
% result = sum(dist)/(mot.nframes+mot.njoints);
% result = result*result;
%tmp=sum(tmp)/mot.nframes;
%tmp=sum(tmp,1)/mot.njoints;
% tmp=(ones(size(tmp))-tmp)*2;
% result(i)=tmp/mot.nframes;
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficients6D.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficients6D.m
| 3,799 |
utf_8
|
d1304249fe489a02f37a81ff7f9de89b
|
% FUNCTION findCoefficients searches for optimal coefficients to
% reconstruct a given motion out of a given tensor. It uses the Matlab
% Optimization Toolbox.
% INPUT:
% Tensor: struct: containing data, core, matrices.
% newMot: struct: the motion that should be reconstructed.
% varargin{1}: string: Type of data Representation 'Quat' or 'Position'
% varargin{2}: cell of arrays: Start values for optimization values
% have to correspond to the number of natural and
% technical modes.
%
% OUTPUT:
% X: cell array: coefficients found for best solution
% Y: cell array: X normalized to length 1
% mot: struct: motion reconstructed with X
% d: float: distance, correponding to the used distance
% measure between mot and newMot.
function [X] = findCoefficients6D(Tensor,newMot,varargin)
% Align first new Motion like all others!
% % [skel,fitmot]=reconstructMotionT(Tensor,[1 1 1]);
% % skel = readASF(Tensor.skeletons{1,1});
% % newMot=fitMotion(skel,newMot);
% % % Timewarp motion
% % [newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=Tensor.numTechnicalModes;
nrOfNaturalModes =Tensor.numNaturalModes;
% Compute mode-n-product of core tensor and all matrices related to
% technical modes
core_tmp=Tensor.core;
for i=1:nrOfTechnicalModes
core_tmp=modeNproduct(core_tmp,Tensor.factors{i},i);
end
root_tmp=Tensor.rootcore;
for i=1:nrOfTechnicalModes-1
root_tmp=modeNproduct(root_tmp,Tensor.rootfactors{i},i);
end
iter=500;
% Set options for optimization
options = optimset('Display','iter','MaxFunEvals',iter*12,'MaxIter',iter,'TolFun',1e-1);
% Set lower and upper bounds for optimization variable x
dimvec=size(core_tmp);
n=sum(dimvec(nrOfTechnicalModes+1:end));
lb=-0.5*ones(1,n);
ub= 1.5*ones(1,n);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
x0=0;
for i=1:nrOfNaturalModes
for j=1:dimvec(nrOfTechnicalModes+i)
x0=[x0 1/dimvec(nrOfTechnicalModes+i)];
end
end
x0=x0(2:end);
setx0=false;
switch nargin
case 2
DataRep='Quat';
case 3
DataRep=varargin{1};
case 4
x0=varargin{2};
setx0=true;
DataRep=varargin{1};
case 5
x0=varargin{2};
setx0=true;
DataRep=varargin{1};
skel=varargin{3};
otherwise
disp('Wrong number of arguments');
end
% if(setx0)
lb= -ones(1,n);
ub= 2*ones(1,n);
% else
% lb=zeros(1,n);
% ub= inf(1,n);
% end
fprintf('\nlower bound x0 upper bound\n');
disp([lb' x0' ub']);
tmpTensor=Tensor;
tmpTensor.core=core_tmp;
tmpTensor.rootcore=root_tmp;
[X,RESNORM,RESIDUAL] = ...
lsqnonlin(@(x) ...
objfun( x,tmpTensor,newMot, ...
nrOfNaturalModes,nrOfTechnicalModes,...
dimvec,skel,DataRep) ...
,x0,lb,ub,options);
% Show computed coefficients
% for i=1:nrOfNaturalModes
% X{i}=Y(1:dimvec(i+nrOfTechnicalModes));
% % X2{i}=round(X{i});
% Y=Y(dimvec(i+nrOfTechnicalModes)+1:size(Y,2));
% end
%
% % Construct motion with computed coefficients and compute mean error
% % [skel ,mot] =constructMotion(Tensor,X,skel,DataRep);
% % [skel2,mot2]=constructMotion(Tensor,X2);
%
% % d =compareMotions(mot, newMot,DataRep);
% % d2=compareMotions(mot2,newMot);
%
% % if d2<d
% % X=X2;
% % d=d2;
% % mot=mot2;
% % end
%
% for i=1:nrOfNaturalModes
% fprintf('\n X{%i}\n',i);
% disp(X{i}');
% end
% fprintf('Mean error of joint orientations: E = %.3f degrees.\n',d);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findCoefficientsModeSA.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/MMM/findCoefficientsModeSA.m
| 1,499 |
utf_8
|
d2e2bcde84272837a3364e6539d0ab3c
|
function [X]=findCoefficientsModeSA(Tensor,newMot)
% Prepare Data for optimization:
% Align first new Motion like all others!
[skel,fitmot]=reconstructMotionT(Tensor,[1 1 1]);
skel = readASF(Tensor.skeletons{1,1});
newMot=fitMotion(skel,newMot);
% Timewarp motion
[newMot]=SimpleDTW(fitmot,skel,newMot);
nrOfTechnicalModes=Tensor.numTechnicalModes;
nrOfNaturalModes=Tensor.numNaturalModes;
% Compute mode-n-product of core tensor and all matrices related to
% technical modes
core_tmp=Tensor.core;
for i=1:nrOfTechnicalModes
core_tmp=modeNproduct(core_tmp,Tensor.factors{i},i);
end
root_tmp=Tensor.rootcore;
for i=1:nrOfTechnicalModes-1
root_tmp=modeNproduct(root_tmp,Tensor.rootfactors{i},i);
end
dimvec=size(core_tmp);
% Define used representation of motion data within the Tensor and
% define starting guess x0 if not set by user (through varargin)
x0=0;
for i=1:nrOfNaturalModes
for j=1:dimvec(nrOfTechnicalModes+i)
x0=[x0 1/dimvec(nrOfTechnicalModes+i)];
end
end
x0=x0(2:end)
X=Array2Cell(x0,nrOfNaturalModes,nrOfTechnicalModes);
D=inf;
time=0;
tic;
while((D>0.1)&&(time<30))
D=rand(1,1);
time=toc;
end
end
function [X]=Array2Cell(x0,nat,tec)
for i=1:nat
X{i}=x0(1:dimvec(i+tec));
x0=x0(dimvec(i+tec)+1:size(x0,2));
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildPCAMatrixFromDir.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/PCA/buildPCAMatrixFromDir.m
| 4,770 |
utf_8
|
297a15d203d30c5ca76eba4a932f4f77
|
function [Matrix] = buildPCAMatrixFromDir(p,varargin)
switch nargin
case 1
maxRep =3;
dataRep='Quat';
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 2
maxRep =varargin{1};
dataRep='Quat';
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 3
maxRep =varargin{1};
dataRep=varargin{2};
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 4
maxRep =varargin{1};
dataRep=varargin{2};
Styles =varargin{3};
otherwise
error('Wrong number of Args');
end
Matrix.DataRep=dataRep;
% define size of representation:
switch dataRep
case 'Quat'
dimDataRep=4;
case 'Position'
dimDataRep=3;
case 'ExpMap'
dimDataRep=3;
case 'Acc'
dimDataRep=3;
otherwise
error('buildPCAMatrixFromDir: Wrong Date specified in var: dataRep');
end
% Check if Backslash is included in path and append extension:
if(p(end)~=filesep)
p=[p filesep];
end
ext='*.amc';
% Check if there is a directory for every style:
numStyles=size(Styles,2);
for s=1:numStyles
if(~exist([p Styles{1,s}],'dir'))
error(['buildTensorStyleActRep_jt: Dir for Style ' Styles{1,s} ' does not exist!']);
end
end
% Get Lists of files:
for s=1:numStyles
listofFiles{s}=dir([p Styles{1,s} filesep ext]);
end
% Known Actors: bd,bk,mm,dg,tr
actors{1}='HDM_bd';
actors{2}='HDM_bk';
actors{3}='HDM_dg';
actors{4}='HDM_mm';
actors{5}='HDM_tr';
numActors=size(actors,2);
for s=1:numStyles
%Count repetitions of each actor
for a=1:numActors
reps(a,s)=countActor(listofFiles{s},actors{a});
end
end
Matrix.data =[];
Matrix.rootdata =[];
for s=1:numStyles
file=1;
for a=1:numActors
for r=1:max(maxRep,reps(a,s))
if (r<=maxRep)
fprintf('Loading motion %i%i%i - ',s,a,r);
skelfile = fullfile(p, Styles{1,s}, [actors{a} '.asf']);
motfile = fullfile(p, Styles{1,s}, listofFiles{s}(file).name);
[skel,mot] = readMocap(skelfile,motfile);
mot = changeFrameRate(skel,mot,30);
mot = fitMotion(skel,mot);
Matrix.rootdata = [Matrix.rootdata mot.rootTranslation];
mot=addAccToMot(mot);
tmpMatrix=zeros(mot.njoints*dimDataRep,mot.nframes);
for joint=1:mot.njoints
% Tensor.joints{joint,s,a,r}=mot.jointNames{joint};
switch dataRep
case 'Quat'
if(~isempty(mot.rotationQuat{joint}))
tmpMatrix(joint*4-3:joint*4,:)=mot.rotationQuat{joint};
else
tmpMatrix(joint*4-3,:) = ones(1,mot.nframes);
tmpMatrix(joint*4-2:joint*4,:)=zeros(3,mot.nframes);
end
case 'Position'
error('Not implemented');
case 'ExpMap'
if(~isempty(mot.rotationQuat{joint}))
tmpMatrix(joint*3-2:joint*3,:)=quatlog(mot.rotationQuat{joint});
else
tmpMatrix(joint*3-2:joint*3,:)=zeros(3,mot.nframes);
end
case 'Acc'
tmpMatrix(joint*3-2:joint*3,:)=mot.jointAccelerations{joint};
otherwise
error('buildTensorStyleActRep_jt: Wrong Type specified in var: dataRep\n');
end
end
Matrix.data=[Matrix.data tmpMatrix];
end
if (r<reps(a,s)||r==max(maxRep,reps(a,s)))
file=file+1;
end
end
end
end
[Matrix.rootcoefs,Matrix.rootscores,Matrix.rootvariances,Matrix.roott2] = princomp(Matrix.rootdata');
[Matrix.coefs,Matrix.scores,Matrix.variances,Matrix.t2] = princomp(Matrix.data');
Matrix.mean =mean(Matrix.data,2);
Matrix.rootmean=mean(Matrix.rootdata,2);
Matrix.cov =cov(Matrix.data');
Matrix.rootCov =cov(Matrix.rootdata);
Matrix.inv =pinv(Matrix.cov');
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
precision_recall_diagram2.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/Retrieval/precision_recall_diagram2.m
| 4,877 |
utf_8
|
0707860911207e2bde51c2b2b4f364d4
|
function [ output_args ] = precision_recall_diagram2( DB_info, mClasses, results1, results2, recompute, m, n, tau1, tau2 )
dbs = dbstack;
fullPath = dbs(1).name(1:max(strfind(dbs(1).name, '\')));
saveFileName = 'precision_recall_diagram_cache';
if ~iscell(mClasses)
motionClasses{1} = mClasses;
else
motionClasses = mClasses;
end
if nargin > 3
if strfind(lower(inputname(3)), 'c3d')
labelTxt1 = 'LCS';
elseif strfind(lower(inputname(3)), 'amc')
labelTxt1 = 'ASF/AMC';
end
if strfind(lower(inputname(4)), 'c3d')
labelTxt2 = 'LCS';
elseif strfind(lower(inputname(4)), 'amc')
labelTxt2 = 'ASF/AMC';
end
end
if nargin < 5
recompute = false;
end
if nargin < 6
b = length(motionClasses);
m = floor(sqrt(b));
if m*m ~= b
m = m + 1;
end
n = double(int32(b/m));
if mod(b,m)~=0
n = n + 1;
end
end
if nargin < 8
tau1 = 0.02;
tau2 = 1;
end
if recompute
for i=1:length(motionClasses)
motionClass = motionClasses{i};
disp(motionClass);
%
% if i < 120
% continue;
% end
idx1 = strmatch(motionClass, {results1.category}', 'exact');
idx2 = strmatch(motionClass, {results2.category}', 'exact');
if isempty(idx1) || isempty(idx2)
error('Motion class not contained in results!');
end
hits1 = results1(idx1).hits;
hits2 = results2(idx2).hits;
% hitIdx11 = find([hits1.cost] <= tau1);
% hitIdx21 = find([hits2.cost] <= tau1);
% hits11 = hits1(hitIdx11);
% hits21 = hits2(hitIdx21);
hits11 = hits1(1:min(20, length(hits1)));
hits21 = hits2(1:min(20, length(hits2)));
if (~isempty(hits1) && ~isempty(hits2))
hitIdx12 = find([hits1.cost] <= tau2);
hitIdx22 = find([hits2.cost] <= tau2);
hits12 = hits1(hitIdx12);
hits22 = hits2(hitIdx22);
else
hits12 = [];
hits22 = [];
end
[precision1{1,i}, recall1{1,i}, n_relevant1{1,i}] = precision_recall2(motionClass, hits11, false, DB_info);
[precision1{2,i}, recall1{2,i}, n_relevant1{2,i}] = precision_recall2(motionClass, hits21, false, DB_info);
[precision2{1,i}, recall2{1,i}, n_relevant2{1,i}] = precision_recall2(motionClass, hits12, false, DB_info);
[precision2{2,i}, recall2{2,i}, n_relevant2{2,i}] = precision_recall2(motionClass, hits22, false, DB_info);
end
save(fullfile(fullPath, 'Cache', saveFileName), 'precision1', 'recall1', 'n_relevant1', 'precision2', 'recall2', 'n_relevant2');
else
load(fullfile(fullPath, 'Cache', saveFileName));
end
figure;
lineColors = {[0 0 1], [1 0 0], [0 0 1], [1 0 0], [0.6 0.6 1], [1 0.6 0.6]};
for i=1:length(motionClasses)
motionClass = motionClasses{i};
h=subplot(m,n,i);
set(h, 'ButtonDownFcn', {@onClick, motionClass, i });
hold on;
plot(recall2{1,i}, precision2{1,i}, 'g');
plot(recall2{2,i}, precision2{2,i}, 'y');
plot(recall1{1,i}, precision1{1,i}, 'g');
plot(recall1{2,i}, precision1{2,i}, 'y');
% delimiters of two tau-curves
plot(recall1{1,i}(end), precision1{1,i}(end), 'gx');
plot(recall1{2,i}(end), precision1{2,i}(end), 'yx');
plotLines = get(gca, 'Children');
for j=1:length(plotLines)
set(plotLines(j), 'Color', lineColors{j});
end
% % delimiters of Top20
% top = min(20, min(length(recall1{1,i}), length(recall1{2,i})));
% topX = recall1{1,i}(top) / max(recall1{1,i}(top), precision1{1,i}(top));
% topY = precision1{1,i}(top) / max(recall1{1,i}(top), precision1{1,i}(top));
% l = line([0 topX], [0 topY]);
% set(l, 'color', [0.6*ones(3,1)]);
% if length(motionClasses)==1
% plot(recall{1,i}, precision{1,i}, 'rx');
% plot(recall{2,i}, precision{2,i}, 'bx');
% xlabel('recall');
% ylabel('precision');
% end
set(gca, 'XLim', [0 1.05]);
set(gca, 'YLim', [0 1.05]);
if length(motionClasses) < 5
legend({labelTxt1, labelTxt2});
else
set(gca, 'XTickLabel', []);
set(gca, 'YTickLabel', []);
t = text(0.5, -0.1, motionClass);
set(t, 'horizontalalignment', 'center');
set(t, 'FontSize', 6);
% t=title(motionClass);
% set(t, 'FontSize', 6);
% set(t, 'Position', [0.55 1.2 1])
end
end
% ---------------------------------------------------
function onClick( src, eventdata, motionClass, number)
set(gcf, 'Name', [motionClass ' (Nr. ' num2str(number) ')']);
disp(motionClass);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
HitsYesNoDifferencePano.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/Retrieval/HitsYesNoDifferencePano.m
| 4,141 |
utf_8
|
b08d69ae3c4bc0effd091b0beb5e558c
|
function HitsYesNoDifferencePano( DB1, DB2, recompute, tau )
% HitsYesNoDifferencePano( DB1, DB2, recompute, tau )
global VARS_GLOBAL
saveFileName = 'hitsYesNoDifference_cache';
if nargin < 2
DB1 = 'HDM05_cut_c3d_dipl';
DB2 = 'HDM05_cut_amc_dipl';
end
if nargin < 3
recompute = false;
end
if nargin < 4
% tau = 0.01;
tau = 1;
end
dbs = dbstack;
fullPath = dbs(1).name(1:max(strfind(dbs(1).name, '\')));
load he_motion_classes;
load HE_DB_info;
keyframesThreshLo = 0.1;
par.thresh_lo = keyframesThreshLo;
par.thresh_hi = 1-par.thresh_lo;
load(['retrieval_results_' DB1 '_' DB1 '_' num2str(1000*par.thresh_lo)]);
VARS_GLOBAL.HitsYesNoDifferenceResults1 = results;
load(['retrieval_results_' DB2 '_' DB2 '_' num2str(1000*par.thresh_lo)]);
VARS_GLOBAL.HitsYesNoDifferenceResults2 = results;
clear results;
if recompute
for i=1:length(motion_classes)
motionClass = motion_classes{i};
disp(motionClass);
idx1 = strmatch(motionClass, {VARS_GLOBAL.HitsYesNoDifferenceResults1.category}', 'exact');
idx2 = strmatch(motionClass, {VARS_GLOBAL.HitsYesNoDifferenceResults2.category}', 'exact');
hitIdx1 = find([VARS_GLOBAL.HitsYesNoDifferenceResults1(idx1).hits.cost] < tau);
hitIdx2 = find([VARS_GLOBAL.HitsYesNoDifferenceResults2(idx2).hits.cost] < tau);
if isempty(hitIdx1) || isempty(hitIdx2)
error('Motion class not contained in results!');
end
hits1 = VARS_GLOBAL.HitsYesNoDifferenceResults1(idx1).hits(hitIdx1);
hits2 = VARS_GLOBAL.HitsYesNoDifferenceResults2(idx2).hits(hitIdx2);
[precision{1,i}, recall{1,i}, n_relevant{1,i}] = precision_recall(motionClass, hits1, false, DB_info);
[precision{2,i}, recall{2,i}, n_relevant{2,i}] = precision_recall(motionClass, hits2, false, DB_info);
end
save(fullfile(fullPath, 'Cache', saveFileName), 'recall', 'precision', 'n_relevant');
else
load(fullfile(fullPath, 'Cache', saveFileName));
end
for i=1:length(motion_classes)
hitYesNo1(i,1:length(recall{1,i})) = diff([0 recall{1,i}*n_relevant{1,i}]);
hitYesNo2(i,1:length(recall{2,i})) = diff([0 recall{2,i}*n_relevant{2,i}]);
% hitYesNo1(i,1:length(recall2{1,i})) = diff([0 recall2{1,i}*n_relevant2{1,i}]);
% hitYesNo2(i,1:length(recall2{2,i})) = diff([0 recall2{2,i}*n_relevant2{2,i}]);
end
range = [1:min(min(size(hitYesNo1,2), size(hitYesNo2,2)), 50)];
% figure;
% imagesc(hitYesNo1(:, range));
% title(DB1, 'Interpreter', 'none');
% colormap(hot);
%
% figure;
% imagesc(hitYesNo2(:, range));
% title(DB2, 'Interpreter', 'none');
% colormap(hot);
figure;
imagesc(hitYesNo1(:, range)' - hitYesNo2(:, range)');
colormap([0 0 1; 1 1 1; 1 0 0]);
h=ylabel('Position');
set(h, 'fontsize', 7);
for i=1:length(motion_classes)
h = text(i, length(range)*1.05, motion_classes{i});
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Rotation', 55);
set(h, 'FontSize', 6);
set(h, 'ButtonDownFcn', @motionClassTextOnClick);
% set(h, 'ButtonDownFcn', {@motionClassTextOnClick, results1, results2});
end
set(gca, 'Position', [0.1 0.3 0.85 0.65]);
set(gcf, 'Position', [236 417 712 413]);
set(gca, 'xticklabel', [])
set(gca, 'fontsize', 7);
% -----------------------------------------------------------
function motionClassTextOnClick(src, eventdata)
global VARS_GLOBAL
t = get(gcf,'selectionType');
motionClass = get(src, 'string');
resultIdx1 = strmatch(motionClass, {VARS_GLOBAL.HitsYesNoDifferenceResults1.category}, 'exact');
resultIdx2 = strmatch(motionClass, {VARS_GLOBAL.HitsYesNoDifferenceResults2.category}, 'exact');
resultBrowser(motionClass);
set(gcf, 'position', [ 5 231 560 650]);
showDeltaDiff2( VARS_GLOBAL.HitsYesNoDifferenceResults2(resultIdx2).hits, VARS_GLOBAL.HitsYesNoDifferenceResults1(resultIdx1).hits, motionClass);
set(gcf, 'position', [ 572 8 560 446]);
showDeltaDiff2( VARS_GLOBAL.HitsYesNoDifferenceResults1(resultIdx1).hits, VARS_GLOBAL.HitsYesNoDifferenceResults2(resultIdx2).hits, motionClass);
set(gcf, 'position', [ 572 462 560 420]);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
compareJoint.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/3DTraj/compareJoint.m
| 2,254 |
utf_8
|
63044f3afb107cbf239a154b0b9d5ef4
|
function compareJoint( skel1, mot1, skel2, mot2, jointName )
% compareJoint( skel1, mot1, skel2, mot2, jointName )
%
% creates a clickable figure showing the l2-distance of the joint
% given by "jointName".
if nargin < 5
help compareJoint
return
end
if length(mot1.jointTrajectories{1}) ~= length(mot2.jointTrajectories{1})
error('Files have different number of frames and cannot be compared!');
end
idx1 = trajectoryID(mot1, jointName);
idx2 = trajectoryID(mot2, jointName);
diff = mot1.jointTrajectories{idx1} - mot2.jointTrajectories{idx2};
diff = sqrt(dot(diff, diff));
h=figure;
set(h, 'Name', mot1.filename);
plot(diff);
set(gca, 'ButtonDownFcn', {@animateOnClick, skel1, mot1, skel2, mot2});
hold on;
meanDiff = mean(diff);
textXPos = mot1.nframes / 20;
plot(meanDiff * ones(1, length(diff)), ':');
plot( (meanDiff-std(diff)) * ones(1, length(diff)), 'r:');
plot( (meanDiff+std(diff)) * ones(1, length(diff)), 'r:');
text( textXPos, meanDiff, 'mean', 'BackgroundColor',[.9 .9 .9]);
text( textXPos, meanDiff-std(diff), 'mean - std', 'BackgroundColor',[.9 .9 .9]);
t=text( textXPos, meanDiff+std(diff), 'mean + std', 'BackgroundColor',[.9 .9 .9]);
xlabel('frames');
ylabel('deviation');
title(['distance between ' upper(jointName) ' trajectories ( std.dev.=' num2str(std(diff)) ')' ], 'Interpreter', 'none');
axis tight;
xlims = get(gca, 'xlim');
axis auto;
set(gca, 'xlim', xlims);
return;
% -------------------------------------------------------------------------
function animateOnClick(varargin)
skel1 = varargin{3};
mot1 = varargin{4};
skel2 = varargin{5};
mot2 = varargin{6};
t = get(gcf,'selectionType');
% try to find animation window
titleText = 'compareJoint animation figure';
children = get(0, 'Children');
animationWindow = [];
for i=1:length(children)
if strcmpi( titleText, get(children(i), 'Name') )
animationWindow = children(i);
end
end
if isempty(animationWindow)
h=figure;
set(h, 'Name', titleText);
else
set(0, 'CurrentFigure', animationWindow);
end
if strcmpi(t, 'alt') % right click
animate(skel2, mot2, 1, 0.5);
elseif strcmpi(t, 'extend') % middle click
animate([skel1, skel2], [mot1, mot2], 1, 0.5);
else
animate(skel1, mot1, 1, 0.5);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
matrix_comparison_similarity.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/FM_MT/matrix_comparison_similarity.m
| 2,704 |
utf_8
|
24f1aa0b2da611ae524fe0e4e05b3b15
|
function similarity = matrix_comparison_similarity( compFunction, aggrFunction, showDetailsCategory, recomputeMatrix )
% matrix = matrix_comparison_similarity( compFunction, aggrFunction, showDetailsCategory, constBoneLengths, recomputeMatrix )
%
% Shoes similarity of motion classes referring to the given compFunction.
%
% compFunction: 'adv' - advanced bitwise difference (not taking differences in run lengths into account)
% 'bitDiff' - bitwise difference
% 'featureCurves' - std. dev. of difference of feature curves
% 'MTdiff' - difference between Motion-Templates
% aggrFunction: 'mean' - aggregates by calculating the mean value of each category
% 'max' - aggregates by taking the maximum value of each category
% showDetailsCategory : determines right-click behaviour
% constBoneLengths : Enforce constant bonelengths after joint position estimation. Default = true.
% recomputeMatrix : Enforces recomputation of matrix
if nargin < 1
help matrix_comparison
return;
else
if nargin < 2
aggrFunction = 'mean';
end
if nargin < 3
showDetailsCategory = true;
end
if nargin < 4
constBoneLengths = true;
end
if nargin < 5
recomputeMatrix = false;
end
end
load HE_motion_classes;
matrix = matrix_comparison(compFunction, aggrFunction, 0, showDetailsCategory, recomputeMatrix );
for i=1:61
for j=1:61
similarity(i,j) = sum(abs(matrix(:,i) - matrix(:,j)));
end
end
figure;
imagesc(similarity, 'buttondownfcn', {@matrixOnClick, similarity, motion_classes});
axis off;
colormap('hot');
set(colorbar, 'Position', [0.935 0.251 0.02 0.7]);
set(colorbar, 'Fontsize', 7);
for i=1:length(motion_classes)
h = text(-1, i, motion_classes{i});
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Interpreter', 'None');
set(h, 'FontSize', 5);
end
for i=1:length(motion_classes)
h = text(i, length(motion_classes)+1.5, motion_classes{i});
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Rotation', 55);
set(h, 'FontSize', 5);
end
set(gca, 'Position', [0.22 0.25 0.7 0.7]);
% ------------------------------
function matrixOnClick(src, eventdata, similarity, motion_classes)
pointClicked = get(get(src, 'Parent'), 'CurrentPoint');
x = round(pointClicked(1,1));
y = round(pointClicked(1,2));
title([motion_classes{x} ' - ' motion_classes{y} ' ( ' num2str(similarity(x,y)) ' )']);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
matrix_comparison_showFeature.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/FM_MT/matrix_comparison_showFeature.m
| 3,408 |
utf_8
|
f4f93990c3e2efac29ba05d8100f7575
|
function matrix = matrix_comparison_showFeature( src, eventdata, compFunction, diffPerFeature, featureNames, motionClasses, dbName1, dbName2, constBoneLengths, CLIM )
pointClicked = get(get(src, 'Parent'), 'CurrentPoint');
x = round(pointClicked(1,1));
y = round(pointClicked(1,2));
% cmap = hot;
% cmap(1,:) = [0.4 0.4 0.4]; % for not used entries
if x <= length(motionClasses)
motionClass = motionClasses{ x };
% left click and right click: Change title to display information
if y <= length(featureNames)
featureName = featureNames{ y };
set(get(get(src, 'Parent'), 'Title'), 'String', [featureName ' (Nr. ' num2str(y) ') - ' motionClass]);
set(get(get(src, 'Parent'), 'Title'), 'Interpreter', 'none');
end
disp(get(get(gca, 'title'), 'string'));
% right click: Open up comparison for selected motion class
t = get(gcf,'selectionType');
if strcmpi(t, 'alt') % right click
% matrix = zeros(length(featureNames), length(motionClasses));
for i=1:length(motionClasses)
matrix(1:length(diffPerFeature{i}(y,:)),i) = diffPerFeature{i}(y,:)';
end
figure;
% imagesc(matrix, 'buttondownfcn',{@matrix_comparison_showMatrixEntryFeature, y, motionClasses, dbName1, dbName2, constBoneLengths });
if nargin < 10 || isempty(CLIM)
imagesc(matrix);
else
imagesc(matrix, CLIM);
end
set(get(gca, 'Children'), 'buttondownfcn',{@matrix_comparison_showMatrixEntryFeature, y, motionClasses, dbName1, dbName2, constBoneLengths });
axis off;
colormap('hot');
for i=1:length(motionClasses)
rectY = length(diffPerFeature{i}(y,:));
rectHeight = size(matrix, 2) - rectY;
% if rectHeight > 0
h=rectangle('Position', [i-0.5 rectY+0.5 1.02 rectHeight]);
set(h, 'EdgeColor', 'none');
set(h, 'FaceColor', [0.5 0.5 0.5]);
for j=1:rectHeight
h=line([i-0.5 i+0.5], [rectY-0.5+j rectY+0.505+j]);
set(h, 'Color', [0 0 0]);
set(h, 'LineWidth', 0.2);
end
end
set(gcf, 'Name', [featureName ' (Nr. ' num2str(y) ')']);
h = text(-1, floor(size(matrix,1)/2), 'files in motion class');
set(h, 'HorizontalAlignment', 'Center');
set(h, 'Rotation', 90);
set(h, 'FontSize', 8);
for i=1:length(motionClasses)
h = text(i, size(matrix, 1)+1.5, motionClasses{i});
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Rotation', 55);
set(h, 'FontSize', 5);
end
set(gca, 'Position', [0.07 0.155 0.92 0.8]);
set(gcf, 'Position', [4 57 831 646]);
end
else
if y <= length(featureNames)
featureName = featureNames{ y };
set(get(get(src, 'Parent'), 'Title'), 'String', featureName);
else
set(get(get(src, 'Parent'), 'Title'), 'String', 'CLICK IMAGE FOR DETAILS!');
end
end
% -----------------------------------------------------------
function featureTextOnClick(src, eventdata)
fullFeatureName = ['feature_AK_bool_' get(src, 'String') '_robust'];
open(fullFeatureName);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
matrix_comparison.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/FM_MT/matrix_comparison.m
| 5,576 |
utf_8
|
8921b05996863ad882e7c58a6260dab9
|
function [matrix, diffPerFeature] = matrix_comparison( compFunction, aggrFunction, thresh, showDetailsCategory, recomputeMatrix, fontSize )
% [matrix, diffPerFeature] = matrix_comparison( compFunction, aggrFunction, thresh, showDetailsCategory, recomputeMatrix, fontSize )
%
% compFunction: 'adv' - advanced bitwise difference (not taking differences in run lengths into account)
% 'bitDiff' - bitwise difference
% 'featureCurves' - std. dev. of difference of feature curves
% 'MTdiff' - difference between Motion-Templates
% aggrFunction: 'mean' - aggregates by calculating the mean value of each category
% 'max' - aggregates by taking the maximum value of each category
% thresh : cuts matrix values below thresh to zero. Used as multiple of max(max(matrix)).
% showDetailsCategory : determines right-click behaviour
% recomputeMatrix : Enforces recomputation of matrix
% fontSize : xLabel and yLabel font size
if nargin < 1
help matrix_comparison
return;
else
switch lower(compFunction)
case {'adv', 'bitdiff', 'featurecurves', 'featurecurvesadv', 'mtdiff'}
otherwise
error('Unknown compFunction!');
end
if nargin < 2
aggrFunction = 'mean';
end
switch lower(aggrFunction)
case {'mean', 'max'}
otherwise
error('Unknown aggrFunction!');
end
if nargin < 3
thresh = 0;
end
if nargin < 4
showDetailsCategory = true;
end
if nargin < 5
recomputeMatrix = false;
end
if nargin < 6
fontSize = 6;
end
end
dbs = dbstack;
fullPath = dbs(1).name(1:max(strfind(dbs(1).name, '\')));
load all_motion_classes;
categories = motion_classes;
feature_names = getFeatureNames;
global VARS_GLOBAL
dbPath = VARS_GLOBAL.dir_root;
dbName1 = 'HDM05_cut_amc';
dbName2 = 'HDM05_cut_c3d';
% filenames for load + save
matrixFilename = 'matrix';
diffFilename = 'diff';
if ( strcmpi(compFunction, 'mtdiff') || strcmpi(compFunction, 'mtDTW') )
[diff, matrix] = motionTemplateComparison( compFunction, dbName1, dbName2, motion_classes );
diffPerFeature = [];
else
if recomputeMatrix
for i=1:length(categories)
[diff{i}, diffPerFeature{i}] = feature_comparison_category( compFunction, categories{i}, dbName1, dbName2 );
% matrix(:,i) = mean(diffPerFeature{i}, 2);
switch lower(aggrFunction)
case 'mean'
matrix(:,i) = mean(diffPerFeature{i}, 2);
case 'max'
matrix(:,i) = max(diffPerFeature{i}, [], 2);
end
end
save( fullfile(fullPath, 'Cache', [matrixFilename '_' aggrFunction '_' compFunction]), 'matrix');
save( fullfile(fullPath, 'Cache', [diffFilename '_' aggrFunction '_' compFunction]), 'diffPerFeature');
else
load( fullfile(fullPath, 'Cache', [matrixFilename '_' aggrFunction '_' compFunction]) );
load( fullfile(fullPath, 'Cache', [diffFilename '_' aggrFunction '_' compFunction]) );
end
end
if thresh > 0
matrix(find(matrix < thresh*max(max(matrix)))) = 0;
end
h=figure;
set(h, 'Name', [aggrFunction ' of ' compFunction]);
if strcmpi(compFunction, 'mtdiff')
imagesc(matrix, 'buttondownfcn',{@mtDiffOnClick, categories, dbName1, dbName2})
else
if showDetailsCategory
imagesc(matrix, 'buttondownfcn',{@matrix_comparison_showCategory, ...
compFunction, diffPerFeature, feature_names, categories, fullfile(dbPath, dbName1), fullfile(dbPath, dbName2), [0 max(max(cell2mat(diffPerFeature)))] })
else
imagesc(matrix, 'buttondownfcn',{@matrix_comparison_showFeature, ...
compFunction, diffPerFeature, feature_names, categories, fullfile(dbPath, dbName1), fullfile(dbPath, dbName2), [0 max(max(cell2mat(diffPerFeature)))] })
end
end
axis off;
colormap('hot');
for i=1:length(feature_names)
h = text(-1, i, feature_names{i}(17:end-7));
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Interpreter', 'None');
set(h, 'FontSize', fontSize);
set(h, 'ButtonDownFcn', @featureTextOnClick);
end
for i=1:length(categories)
h = text(i, length(feature_names)+1.5, categories{i});
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Rotation', 55);
set(h, 'FontSize', fontSize-1);
end
% set(gcf, 'Position', [620 280 650 580])
children = get(gcf, 'Children');
set(children(1), 'YLim', [0 length(feature_names)+3.7]);
h = colorbar;
% pos = get(h,'Position');
% pos(3)=pos(3)/3;
% pos(1)=0.9;
set(h,'Position',[0.93 0.24 0.02 0.72]);
set(h, 'Fontsize', 7);
set(gca, 'Position', [0.2 0.18 0.72 0.79]);
% -----------------------------------------------------------
function featureTextOnClick(src, eventdata)
fullFeatureName = ['feature_AK_bool_' get(src, 'String') '_robust'];
open(fullFeatureName);
function mtDiffOnClick(src, eventdata, categories, dbName1, dbName2)
pointClicked = get(get(src, 'Parent'), 'CurrentPoint');
x = round(pointClicked(1,1));
y = round(pointClicked(1,2));
t = get(gcf,'selectionType');
if strcmpi(t, 'alt') % right click
% showTemplateComparison( categories{x}, dbName1, dbName2, true, false);
showTemplateComparison( categories{x}, dbName1, dbName2, true, true);
else
title(categories{x});
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
matrix_comparison_showCategory.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/analytics/FM_MT/matrix_comparison_showCategory.m
| 3,289 |
utf_8
|
77817382a114de20eea286d077a390ad
|
function [ output_args ] = matrix_comparison_showCategory( src, eventdata, compFunction, diffPerFeature, featureNames, motionClasses, dbName1, dbName2, CLIM )
pointClicked = get(get(src, 'Parent'), 'CurrentPoint');
x = round(pointClicked(1,1));
y = round(pointClicked(1,2));
if x <= length(motionClasses)
motionClass = motionClasses{ x };
% left click and right click: Change title to display information
if y <= length(featureNames)
featureName = featureNames{ y };
t = title([featureName ' - ' motionClass]);
set(t, 'Interpreter', 'none');
else
t = title(motionClass);
set(t, 'Interpreter', 'none');
end
disp(get(get(gca, 'title'), 'string'));
% right click: Open up comparison for selected motion class
t = get(gcf,'selectionType');
if strcmpi(t, 'alt') % right click
motionClass = motionClasses{x};
% showTemplateComparison(motionClass);
dir1 = fullfile(dbName1, motionClass, filesep);
files1 = dir(fullfile(dir1, '*.c3d'));
dir2 = fullfile(dbName2, motionClass, filesep);
files2 = dir(fullfile(dir2, '*.c3d'));
if isempty(files1) % db1 is AMC, db2 is C3D
files1 = dir(fullfile(dir1, '*.amc'));
else % db1 is C3D, db2 is AMC
files2 = dir(fullfile(dir2, '*.amc'));
end
maxFiles = min([length(files1) length(files2)]);
files1 = strcat(mat2cell(repmat(dir1,maxFiles,1), ones(maxFiles,1)), {files1(1:maxFiles).name}');
files2 = strcat(mat2cell(repmat(dir2,maxFiles,1), ones(maxFiles,1)), {files2(1:maxFiles).name}');
figure;
if nargin < 10 || isempty(CLIM)
imagesc(diffPerFeature{x});
else
imagesc(diffPerFeature{x}, CLIM);
end
set(get(gca, 'Children'), 'buttondownfcn',{@matrix_comparison_showMatrixEntry, featureNames, motionClass, files1, files2 });
axis off;
colormap('hot');
set(gca, 'Position', [0.25 0.11 0.72 0.85]);
h = text(floor(length(files1)/2), length(featureNames)+2, ['files in motion class ' motionClass]);
set(h, 'HorizontalAlignment', 'Center');
for i=1:length(featureNames)
h = text(0, i, featureNames{i}(17:end-7));
set(h, 'HorizontalAlignment', 'Right');
set(h, 'Interpreter', 'None');
set(h, 'FontSize', 8);
set(h, 'ButtonDownFcn', @featureTextOnClick);
end
set(gcf, 'Position', [373 58 650 631]);
end
else
if y <= length(featureNames)
featureName = featureNames{ y };
set(get(get(src, 'Parent'), 'Title'), 'String', featureName);
else
set(get(get(src, 'Parent'), 'Title'), 'String', 'CLICK IMAGE FOR DETAILS!');
end
end
% -----------------------------------------------------------
function featureTextOnClick(src, eventdata)
t = get(gcf,'selectionType');
if strcmpi(t, 'alt') % right click
fullFeatureName = ['feature_AK_bool_' get(src, 'String') '_robust'];
open(fullFeatureName);
else
disp(get(src, 'String'));
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
new_motionTemplateGenerateReal_realInputWeighted.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/motion_templates/new_motionTemplateGenerateReal_realInputWeighted.m
| 7,916 |
utf_8
|
a136948ee28935380adc7ee969f800ac
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Generation of motion template
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function [Utemplate,UtemplateWeights,Vcost] = new_motionTemplateGenerateReal_realInputWeighted(U,Uweights,parameter)
% function [Utemplate,UtemplateWeights,Vcost] = new_motionTemplateGenerateReal_realInputWeighted(U,parameter,Uweights_in)
numU = size(U,1);
dimU = size(U{1},1);
lenV = size(U{1},2);
if (parameter.VrepWoverlapFactor<=0)
parameter.VrepWoverlapFactor = 1/(numU-1);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% DTW computatation: U{1} with U{2},...,U{numU}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
VrepW = cell(numU,1);
Vmatch = cell(numU,1);
Vcost = zeros(numU,1);
UwarpWeights = zeros(numU,lenV);
UwarpWeights(1,:) = Uweights{1};
Uwarp = cell(numU,1);
Uwarp{1}=U{1};
for k=2:numU
[Uwarp{k},UwarpWeights(k,:),VrepW{k},Vmatch{k},Vcost(k)] = new_motionTemplateDTWReal_realInputWeighted(U{1},Uweights{1},U{k},Uweights{k},parameter);
end
if (parameter.templateComputationStrategy == 6) || (parameter.templateComputationStrategy == 8)
% only idendical values are kept. Everythind else is set to 0.5
UtemplateHelp = Uwarp{1};
for k=2:numU
differences = find(UtemplateHelp ~= Uwarp{k});
UtemplateHelp(differences) = 0.5*ones(size(differences));
end
%finally average all weights.
UtemplateHelpWeights = sum(UwarpWeights,1);
UtemplateHelpWeights = UtemplateHelpWeights/numU;
else %use real averaging
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Utemplate
% Compute the weighted average over all templates that were generated in the
% preceding step. Note that all templates have the length of the reference
% stream V==U{1}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
UtemplateHelp = zeros(size(U{1}));
UtemplateHelpWeights = sum(UwarpWeights,1);
for k=1:numU
X = repmat(UwarpWeights(k,:),dimU,1);
UtemplateHelp = UtemplateHelp + X.*Uwarp{k};
end
X = repmat(UtemplateHelpWeights,dimU,1);
UtemplateHelp = UtemplateHelp./X;
UtemplateHelpWeights = UtemplateHelpWeights/numU;
end
if (parameter.conjoin == 0)
Utemplate = UtemplateHelp;
UtemplateWeights = UtemplateHelpWeights;
else
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% VrepWConjoin
% All data streams U_i for i=2:numU have been aligned with the reference stream
% V==U_1. Multiple matching frames from a U_i to a frame in V have been
% recorded in the VrepW{i}, each entry of which represents a segment in V.
% Now find the "connected components" of these segments within V.
% That is, find longest contiguous segments within V such that there is a
% covering with overlapping segments from the VrepW{i}, where two
% segments A and B are "overlapping" if there is a sequence of segments that
% connects A and B through simple overlap relations.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
VrepWconcat = VrepW{2};
for k=3:numU
VrepWconcat = [VrepWconcat VrepW{k}];
end
num = size(VrepWconcat,2);
VrepWconjoin = zeros(2,num);
[a,b] = sort(VrepWconcat(1,:));
VrepWconcat = VrepWconcat(:,b);
X = VrepWconcat;
B = zeros(size(X,2),max(max(VrepWconcat)));
s = (numU-1)*ones(size(B,2),1);
s(1) = 0;
p = X(1,1);
for k=1:size(X,2)
B(k,X(1,k):X(2,k)) = 1;
if (X(1,k)~=p) % this code will execute each time a new block of start indices begins
p = X(1,k);
s(p) = sum(B(1:k-1,p));
end
end
% visualization %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if parameter.visVrepW == 1
%X = VrepWconcat(:,(VrepWconcat(1,:)~=VrepWconcat(2,:)));
figure;
subplot(2,2,1);
xlabel('VrepWconcat');
imagesc(B,[0 1]);
colormap hot;
subplot(2,2,2);
xlabel('Column sums over VrepWconcat');
b = sum(B);
plot(b);
set(gca,'xlim',[1 length(b)]);
subplot(2,2,3);
xlabel('Upward column sums at lower contour');
plot(s);
line([1 length(b)],[parameter.VrepWoverlapFactor*(numU-1) parameter.VrepWoverlapFactor*(numU-1)],'color','r');
set(gca,'xlim',[1 length(b)],'ylim',[0 numU]);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% counter = 0;
% k = 0;
% while k<num
% k = k+1;
% pos1 = VrepWconcat(1,k);
% pos2 = VrepWconcat(2,k);
% while k<num && VrepWconcat(1,k+1)<=pos2
% k = k+1;
% pos2 = max(pos2,VrepWconcat(2,k));
% end
% counter = counter+1;
% VrepWconjoin(1,counter)=pos1;
% VrepWconjoin(2,counter)=pos2;
% end
% VrepWconjoin = VrepWconjoin(:,1:counter);
v = find(s<parameter.VrepWoverlapFactor*(numU-1))';
w = [v(2:end)-1 size(B,2)];
VrepWconjoin = [v; w];
% visualization %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if parameter.visVrepW == 1
B = zeros(size(VrepWconjoin,2),max(max(VrepWconjoin)));
for k=1:size(VrepWconjoin,2)
B(k,VrepWconjoin(1,k):VrepWconjoin(2,k)) = 1;
end
subplot(2,2,4);
xlabel('VrepWconjoin');
imagesc(B,[0 1]);
colormap hot;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% UtemplateConjoin
% Now that a segmentation of V has been determined (VrepWconjoin),
% compute the weighted average of the columns within Utemplate corresponding to each
% of these segments, resulting in UtemplateConjoin.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
UtemplateWeights = zeros(1,size(VrepWconjoin,2));
Utemplate = zeros(dimU,size(VrepWconjoin,2));
for k=1:size(VrepWconjoin,2)
X = repmat(UtemplateHelpWeights(VrepWconjoin(1,k):VrepWconjoin(2,k)),dimU,1);
pattern = sum(X.*UtemplateHelp(:,VrepWconjoin(1,k):VrepWconjoin(2,k)),2);
UtemplateWeights(k) = UtemplateWeights(k) + sum(UtemplateHelpWeights(VrepWconjoin(1,k):VrepWconjoin(2,k)));
pattern = pattern/UtemplateWeights(k);
Utemplate(:,k) = pattern;
end
% for k=1:size(VrepWconjoin,2)
% pattern = sum(UtemplateHelp(:,VrepWconjoin(1,k):VrepWconjoin(2,k)),2);
% UtemplateWeights(k) = UtemplateWeights(k) + sum(UtemplateHelpWeights(VrepWconjoin(1,k):VrepWconjoin(2,k)));
% pattern = pattern/(VrepWconjoin(2,k)-VrepWconjoin(1,k)+1);
% Utemplate(:,k) = pattern;
% end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Visualization
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if parameter.visTemplate == 1
figure;
set(gcf,'Position',[55 70 560 861]);
subplot(6,1,1);
plot(UtemplateHelpWeights);
title('UtemplateHelp');
colorbar;
subplot(6,1,2);
colormap(hot);
imagesc(UtemplateHelp,[0 1]);
colorbar;
subplot(6,1,3);
plot(UtemplateWeights);
title('Utemplate');
colorbar;
subplot(6,1,4);
colormap(hot);
imagesc(Utemplate,[0 1]);
colorbar;
drawnow;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
new_motionTemplateDTWReal_realInputWeighted.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/motion_templates/new_motionTemplateDTWReal_realInputWeighted.m
| 5,156 |
utf_8
|
9d7cbd47ace8c08991cdf6d226654402
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Cost of features
% by Meinard Mueller, 03.06.2005
%
% V n times p matrix, data stream of length n with p dimensional feature vectors
% W m times p matrix, data stream of length n with p dimensional feature vectors
% match num x 2 array encoding the matched indices
% costF 1 x p array containing the costs per feature
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function [Wwarp,WwarpWeights,VrepW,match,cost,C,D]=motionTemplateDTW_realInputWeighted(V,Vweights,W,Wweights,parameter)
V = V';
W = W';
n = size(V,1);
p = size(V,2);
m = size(W,1);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Global cost matrix for matching i-th feature with j-th feature
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% C = zeros(n,m);
% for i=1:n
% for j=1:m
% C(i,j) = norm(V(i,:)-W(j,:),1) / p;
% end
% end
% computes pairwise L_1 distance between feature vectors
%%%%%%%% equivalent C code, improves efficiency by roughly a factor of 10
C = C_DTW_compute_Creal(V,W);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
C = C.*((repmat(Vweights',1,m)+repmat(Wweights,n,1))/2);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Computing optimal match by dynamic programming
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %Attention: MATLAB indexing always begins with index 1
% D = zeros(n,m);
% E = zeros(n,m);
%
% D(1,1) = C(1,1);
% for i=2:n
% D(i,1)=D(i-1,1)+C(i,1);
% end
% for j=2:m
% D(1,j)=D(1,j-1)+C(1,j);
% end
%
% for i=2:n
% for j=2:m
% [val,E(i,j)] = min([D(i-1,j-1), D(i,j-1), D(i-1,j)]); %diag, horz, vert
% D(i,j) = val+C(i,j);
% end
% end
%%%%%%%% equivalent C code, improves efficiency
[D,E] = C_DTW_compute_D(C);
i = n;
j = m;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
match = zeros(n+m-1,2);
k=0;
while ((i>1) & (j>1))
k=k+1;
match(k,:)=[i j];
if E(i,j)==1
j=j-1;
i=i-1;
elseif E(i,j)==2
j=j-1;
else
i=i-1;
end
end
k = k+1;
match(k,:)=[i j];
while (i>1)
i = i-1;
k=k+1;
match(k,:)=[i j];
end
while (j>1)
j = j-1;
k=k+1;
match(k,:)=[i j];
end
match = match([1:k],:);
match = flipud(match);
cost = D(n,m);
match = match';
match_v = match(1,:);
match_w = match(2,:);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% De-warp VwarpTemplate into Vtemplate, yielding a sequence
% of the same length as V, while averaging columns that are
% replicated in the match of V to W using Wweights.
% Normalize each resulting column by the sum of Wweights used
% in the averaging process.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
WwarpMatch = W(match_w,:)';
WwarpMatchWeights = Wweights(match_w);
match_len = length(match_v);
Wwarp = zeros(p,n);
WwarpWeights = zeros(1,n);
[Vruns, Vruns_start, Vrep] = runs_find_constant(match_v);
for k=1:length(Vruns)
b = Vruns(k);
range = Vruns_start(k):Vruns_start(k)+Vrep(k)-1;
X = repmat(WwarpMatchWeights(range),p,1);
Wwarp(:,b) = Wwarp(:,Vruns(k))+sum(X.*WwarpMatch(:,range),2);
WwarpWeights(b) = WwarpWeights(b)+sum(WwarpMatchWeights(range));
end
Wwarp = Wwarp./repmat(WwarpWeights,p,1);
% for k=1:match_len
% b = match_v(k);
% Wwarp(:,b) = Wwarp(:,b)+WwarpMatch(:,k)/Vrep(b);
% WwarpWeights(b) = WwarpWeights(b)+WwarpMatchWeights(k);
% end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Compute a 2 x m matrix VrepW, column j of which
% encodes a pair (start_j,end_j). These entries are indices
% into V, and denote segments within V that match to the same
% frame j in W.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[Wruns, Wruns_start, Wrep] = runs_find_constant(match_w);
VrepW = [match_v(Wruns_start); match_v(Wruns_start+Wrep-1)];
for k=1:m % normalize each Wweight(j) by the number of elements in V matching to W(j)
WwarpWeights(VrepW(1,k):VrepW(2,k)) = WwarpWeights(VrepW(1,k):VrepW(2,k))/Wrep(k);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Visualizations
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if parameter.visCost == 1
figure;
imagesc(C,[0 p]);
axis xy;
colormap(hot);
colorbar;
title(['Local cost matrix C, D(n,m)=',num2str(D(n,m)),', #(match)=',num2str(size(match,2))]);
hold on;
plot(match(2,:),match(1,:),'.c');
end
if parameter.visWarp == 1
figure;
subplot(2,1,1);
plot(WwarpWeights); title('WwarpWeights');
colorbar;
subplot(2,1,2);
imagesc(Wwarp,[0 1]); title('Wwarp');
colormap(hot); colorbar;
drawnow;
% subplot(2,2,3);
% imagesc(VwarpEqual,[0 1]); title('VwarpEqual');
% subplot(2,2,4);
% imagesc(Vequal,[0 1]); title('Vequal');
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
motionTemplateDTWRetrievalWeighted.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/motion_templates/motionTemplateDTWRetrievalWeighted.m
| 6,504 |
utf_8
|
220ef1c3e5d7c102dce1db2c81257ac5
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Motion Retrieval via substring DTW based on motion templates
% by Meinard Mueller, 10.11.2005
%
% V n times p matrix, data stream of length n with p dimensional feature vectors
% W m times p matrix, data stream of length n with p dimensional feature vectors
% hits struct array, see below
% costF 1 x p array containing the costs per feature
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function [hits,C,D,Delta]=motionTemplateDTWRetrievalWeighted(V,Vweights,W,Wweights,parameter,varargin)
di = [1 0 1];
dj = [1 1 0];
if (nargin>6)
dj = varargin{2};
end
if (nargin>5)
di = varargin{1};
end
if (nargin>7)
dWeights = varargin{3};
else
dWeights = ones(1,length(di));
end
if (parameter.expandV==1)
[V,Vweights] = expandWeightedVectorSequence(V,Vweights);
end
V = V';
W = W';
n = size(V,1);
p = size(V,2);
m = size(W,1);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Global cost matrix for matching i-th feature with j-th feature
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
V_0 = double(V==0)+double(V==0.5);
V_1 = double(V==1)+double(V==0.5);
C = V_1*W' + V_0*(1-W'); % #(concurrences)
C = p-C; % #(disagreements)
C = C/n; % disagreements per frame in V
%C = C/(p*n); % disagreements per feature and frame in V
%C = C_DTW_compute_C(V,W); % slower
num_V_nonfuzzy = max(1,sum(V~=0.5,2));
C = C.*((repmat(Vweights',1,m)+repmat(Wweights,n,1))./(2*repmat(num_V_nonfuzzy,1,m)));
%C = C.*((repmat(Vweights',1,m)+repmat(Wweights,n,1))/2);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Computing optimal match by dynamic programming
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% tic;
% D = zeros(n,m);
% E = zeros(n,m);
%
% D(1,1) = C(1,1);
% for i=2:n
% D(i,1)=D(i-1,1)+C(i,1);
% end
% for j=2:m
% D(1,j)=C(1,j);
% end
%
% for i=2:n
% for j=2:m
% indices = sub2ind(size(D),max(i-di,1),max(j-dj,1));
% [val,E(i,j)] = min(D(indices));
%
% D(i,j) = val + dWeights(E(i,j))*C(i,j);
% end
% end
%%%%%%%% equivalent C code, improves efficiency by roughly a factor of 150
[D,E] = C_DTWpartial_compute_D_variablePathSteps(C,int32(di),int32(dj),dWeights);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%t=toc; fprintf('%f',t)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Matches
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Delta = D(n,:);
match_numMax = parameter.match_numMax;
match_thresh = parameter.match_thresh;
hits = struct('file_id',0,... % file ID, index into idx.files
'file_name','',...
'cost',0,...
'match',0,... % num x 2 array encoding the matched indices
'frame_first_matched_all',0,... % the first frame of this hit (using a continuous frame count for the entire database)
'frame_last_matched_all',0,... % the last frame of this hit (using a continuous frame count for the entire database)
'frame_first_matched',0,... % the first frame of this hit
'frame_last_matched',0,... % the last frame of this hit
'frame_length',0); % the length of this hit measured in frames
hits = repmat(hits,match_numMax,1);
matches_length = zeros(1,length(Delta)); % this array will indicate the length (in frames) of the match starting at a specific point
Delta_help = Delta;
counter = 0;
[cost,pos] = min(Delta_help);
while counter<match_numMax && cost <= match_thresh
i=n;
j=pos;
match_temp = zeros(max(n,m),2);
k=0;
while ((i>1) & (j>1))
k=k+1;
match_temp(k,:)=[i j];
i2 = i - di(E(i,j));
j2 = j - dj(E(i,j));
i = i2;
j = j2;
end
k = k+1;
match_temp(k,:)=[i j];
while (i>1)
i = i-1;
k=k+1;
match_temp(k,:)=[i j];
end
match_temp = match_temp([1:k],:);
match_temp = flipud(match_temp)';
if (matches_length(match_temp(2,1))>0) % has the starting point of this path been recorded previously?
current_match_length = matches_length(match_temp(2,1));
else % new starting point of a match path; record match.
counter = counter+1;
hits(counter).match = match_temp;
hits(counter).frame_first_matched_all = match_temp(2,1);
hits(counter).frame_last_matched_all = match_temp(2,end);
hits(counter).cost = cost;
current_match_length = match_temp(2,end)-match_temp(2,1)+1;
pufferBackward = ceil(parameter.match_startExclusionBackward*current_match_length);
pufferForward = ceil(parameter.match_startExclusionForward*current_match_length);
ind_start = max(match_temp(2,1)-pufferBackward,1);
ind_end = min(match_temp(2,1)+pufferForward,m);
matches_length(ind_start:ind_end) = repmat(current_match_length,1,ind_end-ind_start+1);
end
pufferBackward = ceil(parameter.match_endExclusionBackward*current_match_length);
pufferForward = ceil(parameter.match_endExclusionForward*current_match_length);
ind_start = max(match_temp(2,end)-pufferBackward,1);
ind_end = min(match_temp(2,end)+pufferForward,m);
Delta_help(ind_start:ind_end)=Inf;
[cost,pos] = min(Delta_help);
end
hits = hits(1:counter);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Visualizations
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if parameter.visCost == 1
figure;
set(gcf,'Position',[30 551 1210 398]);
subplot(2,1,1)
C_vis = C;
C_vis(isnan(C_vis))=max(max(C_vis));
imagesc(C_vis);
axis xy;
colormap(hot);
colorbar;
D_vis = D;
D_vis(isnan(D_vis))=max(max(D_vis));
subplot(2,1,2)
imagesc(D_vis);
axis xy;
colormap(hot);
colorbar;
hold on;
for k=1:length(hits)
plot(hits(k).match(2,:),hits(k).match(1,:),'.c');
end
% h=figure;
% plot(Delta);
% set(h,'position',[0 161 1275 229]);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
MT_DisplayRealtimeData.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/XSensMatlab/MT_DisplayRealtimeData.m
| 22,793 |
utf_8
|
a722757e17506c523bd2757224064a50
|
function MT_DisplayRealtimeData(varargin)
%% MT_DisplayRealtimeData(varargin);
%
% Real-time display of calibrated data or 3D orientation data from an MTi
% or MTx
%
% varargin: Input arguments
% 1. COM-port [integer] to which MT is connected (default = 1 , i.e. COM1)
% 2. Display mode [string], choose to view:
% 'caldata' Calibrated inertial and magnetic data (default)
% 'euler' Orientation output in Euler-angles (see Tech Doc for definition)
% 'quat' Orientation output in Quaternions
% 'matrix' Orientation output in rotation Matrix format
% 3. zoom-level, [1=12 s chart, 2=8 s chart (default), 3=4 s chart]
% 4. update rate, [1=slow, 2=medium (default), 3=fast]
%
% example function call:
% "MT_DisplayRealtimeData(1,'euler',2,3)"
%
% Press key on keyboard while running:
%
% Calibrated Inertial and magnetic data MODE:
% "Q" = Quit
% "P" = Pause/Start real-time plot
% "Z" = Toggle Zoom in/out
% "A" = View only accelerometer data
% "G" = View only rate gyro data
% "M" = View only magnetometer data
% "D" = View all MTi / MTx data (default)
%
% Orientation data output MODE:
% "Q" = Quit
% "P" = Pause/Start real-time plot
% "R" = Reset reference orientation (refer to Tech Doc for details)
% "Z" = Zoom in/out (not applicable for Matrix output mode)
% (it is not possible to switch orientation output mode during run-time)
%
% Xsens Technologies B.V., 2002-2007
% v.2.8.4
% tested to be compatible with MATLAB 2006b.
% The display code in this demo code has been changed due to a bug in MATLAB 7.x
% please refer to:
% http://www.mathworks.com/support/bugreports/details.html?rp=207276
% Check to see if number of input arguments is correct
if nargin>4,
error('too many input arguments')
end
% set default values if needed
defaultValues={1,'caldata',2,1.0,1,1.0};
nonemptyIdx=~cellfun('isempty',varargin);
defaultValues(nonemptyIdx)=varargin(nonemptyIdx);
[COMport,DisplayMode,zoom_level_setting,filterSettings_gain,filterSettings_corr,filterSettings_weight]...
=deal(defaultValues{:});
% set up MTObj
h=actxserver('MotionTracker.FilterComponent');
try
% use try-catch to avoid the need to shutdown MATLAB if m-function fails (e.g. during debugging)
% This is needed because MATLAB does not properly release COM-objects when
% the m-function fails/crasehes and the object is not explicitly released with "delete(h)"
% call MT_SetCOMPort is required, unless using COM 1
h.MT_SetCOMPort(COMport);
% request the Sample Frequency (update rate) of the MTi or MTx
SF = h.MT_GetMotionTrackerSampleFrequency;
% call MT_SetFilterSettings is optional
h.MT_SetFilterSettings(filterSettings_gain,filterSettings_corr,filterSettings_weight);
% init figure plotting variables
% set time scale zoom level
zoom_levels=[12*SF,8*SF,4*SF]; % in seconds
zoom_level=zoom_levels(zoom_level_setting);
OldFigureUserData = [0 0 0]; status = 0; last_t=0;
% default vertical zoom
YLims = [-22 22; -1200 1200; -1.8 1.8; -1 1];
ResetFigure =1; first_time=1;
% create figure
% NOTE, the releasing of MTObj is done in the Figure Close Request function
% "mt_figure_close"
[f_handle, p_handle] = mt_create_figure(OldFigureUserData(3), -1 ,h, YLims, SF,...
zoom_level, OldFigureUserData);
% choose what data to ask from MTObj
if strcmp(DisplayMode,'caldata')
Channels = 10;
% request calibrated inertial and magnetic data
h.MT_SetCalibratedOutput(1);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 0])
elseif strcmp(DisplayMode,'euler')
Channels = 3;
% request orientation data in Euler-angles
h.MT_SetOutputMode(1);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 4])
elseif strcmp(DisplayMode,'quat')
Channels = 4;
% request orientation data in quaternions
h.MT_SetOutputMode(0);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 5])
elseif strcmp(DisplayMode,'matrix')
Channels = 9;
% request orientation data in DCM rotation matrix
h.MT_SetOutputMode(2);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 6])
else
disp('unkown mode....stopping, check variable "Channels"!!')
% clean up, release MTObj
delete(h);
clear h;
end
% That's it!
% MTObj is ready to start processing the data stream from the MTi or MTx
h.MT_StartProcess; % start processing data
% init data variable
d=zeros(1,Channels);
last_d = d;
while ishandle(f_handle) && ishandle(h), % check if all objects are alive
if status ~= 0,
last_d=d(end,:);
end
% retreive data from MTObj object
[d,status,N]=MT_get_data(h, Channels);
% Now the data is available! The rest of the code is only to
% display the data and to make it look nice and smooth on display....
if status==1, % default mode...
% retrieve values from figure
CurrentFigureUserData = get(f_handle,'UserData');
if ResetFigure ==1, % check if figure should be reset
last_t=0; % wrap plot around
figureUserData = get(f_handle,'UserData');
% call local function to (re)setup figure
[f_handle, p_handle] = mt_create_figure(CurrentFigureUserData(3), f_handle,h, YLims, SF, zoom_level, figureUserData);
ResetFigure = 0;
end
% create timebase
t=[last_t:last_t+N]./SF;
last_t=last_t+N;
% check if figures should be reset
if any(CurrentFigureUserData ~= OldFigureUserData), % check if figure UserData changed by KeyPress
ResetFigure =1; % make sure plot is reset
first_time =1; % re-initialize zoom levels too
elseif last_t>zoom_level || first_time==1
ResetFigure =1; % make sure plot is reset
first_time =0;
YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle);
else % other wise --> plot
% plot the data using a local funtion
YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle);
end
OldFigureUserData = CurrentFigureUserData;
elseif status>1,
% MTObj not correctly configured, stopping
[str_out]=MT_return_error(status);
disp(str_out);
disp('MTObj not correctly configured, stopping.....');
break
end % if
end % while
% release MTObj is done on figure close...not here
if ishandle(f_handle), % check to make sure figure is not already gone
close(f_handle)
end
catch % try catch for debugging
% make sure MTObj is released even on error
h.MT_StopProcess;
delete(h);
clear h;
% display some information for tracing errors
disp('Error was catched by try-catch.....MTobj released')
crashInfo = lasterror; % retrieve last error message from MATLAB
disp('Line:')
crashInfo.stack.line
disp('Name:')
crashInfo.stack.name
disp('File:')
crashInfo.stack.file
rethrow(lasterror)
end
% -------------------------------------------------------------------------
% end of function MT_DisplayRealtimeData(varargin);
%% -------------------------------------------------------------------------
% LOCAL FUNCTIONS
% -------------------------------------------------------------------------
function [f_handle, p_handle] = mt_create_figure(type, f_handle, h, YLims,SF, zoom_level, figureUserData)
% local function to create the figure for real time plotting of data.
% accepts plot type information for custom plots
%
% if figure does not yet exist enter -1 in figure_handle input
if ~ishandle(f_handle),
% create figure
f_handle = figure('Name','Real-time display of MTi or MTx data','NumberTitle','off');
end
fontSizeUsed = 12;
axisName = {'X' 'Y' 'Z'};
eulerName = {'Roll' 'Pitch' 'Yaw'};
quatName = {'img w q_0' 'x q_1' 'y q_2' 'z q_3'};
% init
p_handle = zeros(9); a_handle = zeros(9);
switch type
case 0% calibrated inertial and magnetic data(default)
for i=1:9,
subplot(3,3,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(mod(i-1,3)+1,:)]);
grid on;
end
tlh = title(a_handle(1),['Acceleration [m/s^2] (press A)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(7),'time [s]'); ylh = ylabel(a_handle(1),'X_S');
tlh = title(a_handle(2),['Gyro [deg/s] (press G)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(8),'time [s]'); ylh = ylabel(a_handle(4),'Y_S');
tlh = title(a_handle(3),['Magnetometer [a.u.] (press M)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(9),'time [s]'); ylh = ylabel(a_handle(7),'Z_S');
case 1 %'acc' (only accelerometer)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName{i} '_S acceleration [m/s^2]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Accelerometer data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 2 % 'gyr' (only gyroscopes)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName(i) '_S angular rate [deg/s]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Rate gyroscope data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 3 % 'mag' (only magnetometers)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName(i) '_S magnetometer [a.u.]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Magnetometer data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 4 % Euler angles
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[eulerName(i) ' [deg]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Euler angle orientation data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
if figureUserData(2)==0, % try to get nice scales on graphs
set(a_handle(1),'ytick',[-180:45:180]); set(a_handle(2),'ytick',[-90:45:90]); set(a_handle(3),'ytick',[-180:45:180]);
elseif figureUserData==1,
set(a_handle(1),'ytick',[-180:2:180]); set(a_handle(2),'ytick',[-90:2:90]); set(a_handle(3),'ytick',[-180:2:180]);
end
case 5 % Quaternion
for i=1:4,
subplot(4,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),quatName(i)); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Quaternion orientation data q_G_S']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 6 % DCM rotation matrix
p_handle(1)=surf(zeros(4,4),'EraseMode','none'); a_handle(1) = gca;
tlh = title(['Rotation Matrix Output R_G_S - MTi / MTx']); set(tlh,'FontSize',fontSizeUsed,'Color','white');
axis ij;
xlh = xlabel('cols'); ylh_1 = ylabel('rows');
% set(f_handle,'Renderer','OpenGL'); % Use OpenGL renderer for smoother plotting
set(f_handle,'Color','black','Colormap',colormap(hsv));
view(2); shading flat;
set(a_handle(1),'CLim',[-1 1]); cbh = colorbar('vert'); set(cbh,'YColor','white')
end
set(f_handle,'CloseRequestFcn',{@mt_figure_close,h,f_handle});
set(f_handle,'KeyPressFcn',{@mt_figure_keypress,h,f_handle});
set(f_handle,'UserData', figureUserData);
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function mt_figure_keypress(obj, eventdata, h, f_handle)
% local function to handle KeyPress events on figure.
% Is used to (P)ause plot, (Z)oom in/out, (D)efault display,
% (A)ccelerometer only, rate (G)yro only, (M)agnetometer only
%
% envoked when a key is pressed when figure is in focus
in_key=lower(get(f_handle,'CurrentCharacter'));
tmp = get(f_handle,'UserData');
switch in_key
case 'p' % pause to view graph
disp('Paused, press any key to continue...')
pause(0.2);% introduce a slight break, because otherwise 1 keystroke is recorded as multiple
figure(f_handle);% raise figure to foreground
pause; % wait for next key stroke
case 'z' % toggle zoom mode
pause(0.2)
figure(f_handle)
if tmp(2) == 0, % check zoom level
set(f_handle,'UserData',[tmp(1) 1 tmp(3)]); % toggle to next zoom mode
else
set(f_handle,'UserData',[tmp(1) 0 tmp(3)]); % toggle to default zoom mode
end
case 'r' % reset orientation
disp('Resetting heading orientation (boresight)...')
pause(0.2);% introduce a slight break, because otherwise 1 keystroke is recorded as multiple
figure(f_handle);% raise figure to foreground
MT_ResetOrientation(h,0,0); % default reset type 0 = heading, do not save to MTS = 0 (second parameter)
case 'a'
disp('Switching to display only 3D accelerometer data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 1]); % set to accelerometer mode
case 'g'
disp('Switching to display only 3D rate gyroscope data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 2]); % set to gyro mode
case 'm'
disp('Switching to display only 3D magnetometer data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 3]); % set to mag mode
case 'd'
disp('Switching the default display mode, all 9 data streams...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 0]); % set to default mode
case 'q'
disp('Quitting demo MT_DisplayRealtimeData...')
pause(0.2)
figure(f_handle)
close(f_handle)
otherwise
disp('Unknown command option....displaying help data')
disp(' ')
eval('help MT_DisplayRealtimeData')
end
if ishandle(f_handle) % needed to check if figure exists (using Q to quit)
if tmp(3)>3,% If in Orientation output mode, IGNORE any change in PlotMode!!!
tmp_new = get(f_handle,'UserData');
set(f_handle,'UserData',[tmp_new(1:2) tmp(3)]);
end
% reset CurrentCharater to no value...
set(f_handle,'CurrentCharacter',' ');
end
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle)
% local function to plot the data using "low-level" set fucntions for smooth plotting
switch CurrentFigureUserData(3) % check plot type
case 0 %default
if CurrentFigureUserData(2), % check if zoomed
band = [0.8 50 0.1]; % define zoom range
YLims = [min(d(1,1:3))-band(1) max(d(1,1:3))+band(1); min(d(1,4:6)./pi.*180)-band(2) max(d(1,4:6)./pi.*180)+band(2);...
min(d(1,7:9))-band(3) max(d(1,7:9))+band(3)];
else % default values of zoom (full range of MT9)
YLims = [-22 22; -1200 1200; -1.8 1.8];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(4),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(7),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','r','LineWidth',2)
% convert the rate of turn data to deg/s instead of rad/s
set(p_handle(2),'XData',t,'YData',([last_d(1,4) d(:,4)'])./pi.*180,'Color','b','LineWidth',2)
set(p_handle(5),'XData',t,'YData',([last_d(1,5) d(:,5)'])./pi.*180,'Color','g','LineWidth',2)
set(p_handle(8),'XData',t,'YData',([last_d(1,6) d(:,6)'])./pi.*180,'Color','r','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,7) d(:,7)'],'Color','b','LineWidth',2)
set(p_handle(6),'XData',t,'YData',[last_d(1,8) d(:,8)'],'Color','g','LineWidth',2)
set(p_handle(9),'XData',t,'YData',[last_d(1,9) d(:,9)'],'Color','r','LineWidth',2)
case 1 % Only accelerometer
if CurrentFigureUserData(2), % check if zoomed
band = 0.8; % define zoom range (in m/s2)
YLims = [min(d(:,1))-band max(d(:,1))+band; min(d(:,2))-band max(d(:,2))+band; ...
min(d(:,3))-band max(d(:,3))+band];
else % default values of zoom (full range of MT9)
YLims = [-25 25; -25 25; -25 25];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',([last_d(1,3) d(:,3)']),'Color','r','LineWidth',2)
case 2 % Only rate gyros
if CurrentFigureUserData(2), % check if zoomed
band = 50; % define zoom range (in deg/s)
YLims = [min(d(:,4)./pi.*180)-band max(d(:,4)./pi.*180)+band;...
min(d(:,5)./pi.*180)-band max(d(:,5)./pi.*180)+band; min(d(:,6)./pi.*180)-band max(d(:,6)./pi.*180)+band];
else % default values of zoom (full range of MT9)
YLims = [-1200 1200; -1200 1200; -1200 1200];
end
% plot the data
set(p_handle(1),'XData',t,'YData',([last_d(1,4) d(:,4)'])./pi.*180,'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',([last_d(1,5) d(:,5)'])./pi.*180,'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',([last_d(1,6) d(:,6)'])./pi.*180,'Color','r','LineWidth',2)
case 3 % Only magnetometers
if CurrentFigureUserData(2), % check if zoomed
band = 0.1; % define zoom range (in a.u.)
YLims = [min(d(:,7))-band max(d(:,7))+band; min(d(:,8))-band max(d(:,8))+band; min(d(:,9))-band max(d(:,9))+band];
else % default values of zoom (full range of MT9)
YLims = [-2.5 2.5; -2.5 2.5; -2.5 2.5];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,7) d(:,7)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,8) d(:,8)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,9) d(:,9)'],'Color','r','LineWidth',2)
case 4 % Euler angles
if CurrentFigureUserData(2), % zoom UserData
band = 6; % define zoom range (in degrees)
YLims = [min(d(:,1))-band max(d(:,1))+band; min(d(:,2))-band max(d(:,2))+band; ...
min(d(:,3))-band max(d(:,3))+band];
else % default values of zoom (full range of Euler angles)
YLims = [-180 180;-90 90;-180 180];
end
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','r','LineWidth',2)
case 5 % Quaternions
if CurrentFigureUserData(2), % zoom UserData
band = 0.14; % define zoom range
YLims = [min(d(1,1))-band max(d(1,1))+band; min(d(1,2))-band max(d(1,2))+band;...
min(d(1,3))-band max(d(1,3))+band; min(d(1,4))-band max(d(1,4))+band]; % not so useful for quaternions...
else % default values of zoom (full range of Euler angles)
YLims = [-1 1; -1 1; -1 1; -1 1;];
end
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','k','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','b','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','g','LineWidth',2)
set(p_handle(4),'XData',t,'YData',[last_d(1,4) d(:,4)'],'Color','r','LineWidth',2)
case 6 % Matrix
YLims = [0 4]; % not used
% only display latest data available in DCM orientation matrix mode
set(p_handle(1),'cdata',[d(end,1:3); d(end,4:6); d(end,7:9)]');
end % switch
% flush the graphics to screen
drawnow
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function mt_figure_close(obj, eventdata, h, f_handle)
% local function to properly release MTObj when the user kills the figure window.
% release MTObj
MT_StopProcess(h)
delete(h);
clear h;
% kill figure window as requested
delete(f_handle)
% -------------------------------------------------------------------------
% end of function
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
XM_DisplayRealtimeData.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/XSensMatlab/XM_DisplayRealtimeData.m
| 24,551 |
utf_8
|
dd8819c5bd670a9720d47dff072c9c2d
|
function XM_DisplayRealtimeData(varargin)
%% XM_DisplayRealtimeData(varargin);
%
% Real-time display of calibrated data or 3D orientation data from an MTi
% or MTx
%
% varargin: Input arguments
% 1. COM-port [integer] to which Xbus Master is connected (default = 1 , i.e. COM1)
% 2. Display mode [string], choose to view:
% 'caldata' Calibrated inertial and magnetic data (default)
% 'euler' Orientation output in Euler-angles (see Tech Doc for definition)
% 'quat' Orientation output in Quaternions
% 'matrix' Orientation output in rotation Matrix format
% 3. zoom-level, [1=12 s chart, 2=8 s chart (default), 3=4 s chart]
% 4. update rate, [1=slow, 2=medium (default), 3=fast]
%
% example function call:
% "XM_DisplayRealtimeData(1,'euler',2,3)"
%
% Press key on keyboard while running:
%
% Calibrated Inertial and magnetic data MODE:
% "Q" = Quit
% "]" = Switch to display next MTx connected on Xbus
% "P" = Pause/Start real-time plot
% "Z" = Toggle Zoom in/out
% "A" = View only accelerometer data
% "G" = View only rate gyro data
% "M" = View only magnetometer data
% "D" = View all MTi / MTx data (default)
%
% Orientation data output MODE:
% "Q" = Quit
% "P" = Pause/Start real-time plot
% "R" = Reset reference orientation (refer to Tech Doc for details)
% "Z" = Zoom in/out (not applicable for Matrix output mode)
% (it is not possible to switch orientation output mode during run-time)
%
% Xsens Technologies B.V., 2002-2007
% v.2.8.4
% tested to be compatible with MATLAB 2006b.
% The display code in this demo code has been changed due to a bug in MATLAB 7.x
% please refer to:
% http://www.mathworks.com/support/bugreports/details.html?rp=207276
% Check to see if number of input arguments is correct
if nargin>4,
error('too many input arguments')
end
% set default values if needed
defaultValues={4,'caldata',2,1.0,1,1.0};
nonemptyIdx=~cellfun('isempty',varargin);
defaultValues(nonemptyIdx)=varargin(nonemptyIdx);
[COMport,DisplayMode,zoom_level_setting,filterSettings_gain,filterSettings_corr,filterSettings_weight]...
=deal(defaultValues{:});
% set up MTObj
h=actxserver('MotionTracker.FilterComponent');
try
% use try-catch to avoid the need to shutdown MATLAB if m-function fails (e.g. during debugging)
% This is needed because MATLAB does not properly release COM-objects when
% the m-function fails/crasehes and the object is not explicitly released with "delete(h)"
% call XM_SetCOMPort is required, unless using COM 1
h.XM_SetCOMPort(COMport);
% request the Sample Frequency (update rate) of the MTi or MTx
SF = h.XM_GetXbusMasterSampleFrequency;
% Query XM for connected sensors
[num_MTs,DIDs]=h.XM_QueryXbusMasterB;
if num_MTs==0,
disp('Xbus Master not found or no sensors found by Xbus Master...exiting.')
disp('Did you select the correct COM-port?')
delete(h); clear h; close(f_handle)
return
end
for i=1:num_MTs,
% Note: Creating Cell array of device IDs for use in Set functions
% below
MT_IDs{i} = DIDs((i*8 - 7):(i*8));
h.XM_SetFilterSettings(char(MT_IDs(i)),filterSettings_gain,filterSettings_corr, filterSettings_weight);
end
% init figure plotting variables
% set time scale zoom level
zoom_levels=[12*SF,8*SF,4*SF]; % in seconds
zoom_level=zoom_levels(zoom_level_setting);
OldFigureUserData = [0 0 0]; status = 0; last_t=0;
% default vertical zoom
YLims = [-22 22; -1200 1200; -1.8 1.8; -1 1];
ResetFigure =1; first_time=1;
% create figure
% NOTE, the releasing of MTObj is done in the Figure Close Request function
% "mt_figure_close"
[f_handle, p_handle] = xm_create_figure(OldFigureUserData(3), -1 ,h, YLims, SF,...
zoom_level, OldFigureUserData, 1, ' ');
% choose what data to ask from MTObj and store in figure UserData
if strcmp(DisplayMode,'caldata')
Channels = 10;
% request calibrated inertial and magnetic data
h.XM_SetCalibratedOutput(1);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 0])
elseif strcmp(DisplayMode,'euler')
Channels = 3;
% request orientation data in Euler-angles
h.XM_SetOutputMode(1);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 4])
elseif strcmp(DisplayMode,'quat')
Channels = 4;
% request orientation data in quaternions
h.XM_SetOutputMode(0);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 5])
elseif strcmp(DisplayMode,'matrix')
Channels = 9;
% request orientation data in DCM rotation matrix
h.XM_SetOutputMode(2);
% set figure UserData to appropriate values (format:
% [MT_being_plotted, Zoom, PlotType])
set(f_handle,'UserData', [0 0 6])
else
disp('unkown mode....stopping, check variable "Channels"!!')
% clean up, release MTObj
delete(h);
clear h;
end
% That's it!
% MTObj is ready to start processing the data stream from the MTi or MTx
h.XM_StartProcess; % start processing data
% init data variable
d=zeros(1,Channels);
last_d = d;
MT_BeingPlotted = 1;
while ishandle(f_handle) && ishandle(h), % check if all objects are alive
if status ~= 0,
last_d=d(end,:);
end
% retreive data from MTObj object
[d,status,N]=XM_get_data(h, Channels, MT_BeingPlotted);
% Now the data is available! The rest of the code is only to
% display the data and to make it look nice and smooth on display....
if status==1, % default mode...
% retrieve values from figure
CurrentFigureUserData = get(f_handle,'UserData');
if ResetFigure ==1, % check if figure should be reset
last_t=0; % wrap plot around
figureUserData = get(f_handle,'UserData');
% call local function to (re)setup figure
[f_handle, p_handle] = xm_create_figure(CurrentFigureUserData(3), f_handle,h, YLims, SF,...
zoom_level, figureUserData, MT_BeingPlotted, MT_IDs);
ResetFigure = 0;
end
% create timebase
t=[last_t:last_t+N]./SF;
last_t=last_t+N;
% check if to change MTx being plotted
if CurrentFigureUserData(1)
MT_BeingPlotted = mod(MT_BeingPlotted+1,num_MTs);
if MT_BeingPlotted==0,
MT_BeingPlotted = num_MTs;
end
CurrentFigureUserData(1) = 0;% reset next MT toggle
set(f_handle,'UserData', CurrentFigureUserData);
ResetFigure =1; % make sure plot is reset
first_time =1; % re-initialize zoom levels too
% check if figures should be reset
elseif any(CurrentFigureUserData ~= OldFigureUserData), % check if figure UserData changed by KeyPress
ResetFigure =1; % make sure plot is reset
first_time =1; % re-initialize zoom levels too
elseif last_t>zoom_level || first_time==1
ResetFigure =1; % make sure plot is reset
first_time =0;
YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle);
else % other wise --> plot
% plot the data using a local funtion
YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle);
end
OldFigureUserData = CurrentFigureUserData;
elseif status>1,
% MTObj not correctly configured, stopping
[str_out]=MT_return_error(status);
disp(str_out);
disp('MTObj not correctly configured, stopping.....');
break
end % if
end % while
% release MTObj is done on figure close...not here
if ishandle(f_handle), % check to make sure figure is not already gone
close(f_handle)
end
catch % try catch for debugging
% make sure MTObj is released even on error
h.XM_StopProcess;
delete(h);
clear h;
% display some information for tracing errors
disp('Error was catched by try-catch.....MTobj released')
crashInfo = lasterror; % retrieve last error message from MATLAB
disp('Line:')
crashInfo.stack.line
disp('Name:')
crashInfo.stack.name
disp('File:')
crashInfo.stack.file
rethrow(lasterror)
end
% -------------------------------------------------------------------------
% end of function XM_DisplayRealtimeData(varargin);
%% -------------------------------------------------------------------------
% LOCAL FUNCTIONS
% -------------------------------------------------------------------------
function [f_handle, p_handle] = xm_create_figure(type, f_handle, h, YLims,SF, zoom_level, figureUserData, MT_BeingPlotted, MT_IDs)
% local function to create the figure for real time plotting of data.
% accepts plot type information for custom plots
%
% if figure does not yet exist enter -1 in figure_handle input
if ~ishandle(f_handle),
% create figure
f_handle = figure('Name','Real-time display of MTi or MTx data','NumberTitle','off');
end
fontSizeUsed = 12;
axisName = {'X' 'Y' 'Z'};
eulerName = {'Roll' 'Pitch' 'Yaw'};
quatName = {'img w q_0' 'x q_1' 'y q_2' 'z q_3'};
% init
p_handle = zeros(9); a_handle = zeros(9);
figure(f_handle); % raise figure
switch type
case 0% calibrated inertial and magnetic data(default)
for i=1:9,
subplot(3,3,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(mod(i-1,3)+1,:)]);
grid on;
end
tlh = title(a_handle(1),['Acceleration [m/s^2] (press A)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(7),'time [s]'); ylh = ylabel(a_handle(1),'X_S');
tlh = title(a_handle(2),['Gyro [deg/s] (press G)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(8),'time [s]'); ylh = ylabel(a_handle(4),'Y_S');
tlh = title(a_handle(3),['Magnetometer [a.u.] (press M)']); set(tlh,'FontSize',fontSizeUsed); xlh = xlabel(a_handle(9),'time [s]'); ylh = ylabel(a_handle(7),'Z_S');
case 1 %'acc' (only accelerometer)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName{i} '_S acceleration [m/s^2]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Accelerometer data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 2 % 'gyr' (only gyroscopes)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName(i) '_S angular rate [deg/s]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Rate gyroscope data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 3 % 'mag' (only magnetometers)
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[axisName(i) '_S magnetometer [a.u.]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Magnetometer data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 4 % Euler angles
for i=1:3,
subplot(3,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),[eulerName(i) ' [deg]']); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Euler angle orientation data']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
if figureUserData(2)==0, % try to get nice scales on graphs
set(a_handle(1),'ytick',[-180:45:180]); set(a_handle(2),'ytick',[-90:45:90]); set(a_handle(3),'ytick',[-180:45:180]);
elseif figureUserData==1,
set(a_handle(1),'ytick',[-180:2:180]); set(a_handle(2),'ytick',[-90:2:90]); set(a_handle(3),'ytick',[-180:2:180]);
end
case 5 % Quaternion
for i=1:4,
subplot(4,1,i), p_handle(i)=plot(0,0,'EraseMode','none');a_handle(i) = gca; axis(a_handle(i),[0 (zoom_level+1)./SF YLims(i,:)]); grid on;
ylh = ylabel(a_handle(i),quatName(i)); set(ylh,'FontSize',fontSizeUsed-1);
end
tlh = title(a_handle(1),['Quaternion orientation data q_G_S']); set(tlh,'FontSize',fontSizeUsed);
xlh = xlabel('time [s]'); set(xlh,'FontSize',fontSizeUsed);
case 6 % DCM rotation matrix
subplot(1,1,1), p_handle(1)=surf(zeros(4,4),'EraseMode','none'); a_handle(1) = gca;
tlh = title(['Rotation Matrix Output R_G_S - MTi / MTx']); set(tlh,'FontSize',fontSizeUsed,'Color','white');
axis ij;
xlh = xlabel('cols'); ylh_1 = ylabel('rows');
% set(f_handle,'Renderer','OpenGL'); % Use OpenGL renderer for smoother plotting
set(f_handle,'Color','black','Colormap',colormap(hsv));
view(2); shading flat;
set(a_handle(1),'CLim',[-1 1]); cbh = colorbar('vert'); set(cbh,'YColor','white')
end
% add text about which MTx data is displayed from
% text will be displayed wrt last axis active...
vertPos = 1.25* min(get(gca,'Ylim'))- 0.25 * max(get(gca,'YLim'));
txtMT_DID = text(0,vertPos,['MTx number ' num2str(MT_BeingPlotted) ': MTx DID = ' char(MT_IDs(MT_BeingPlotted))]);
set(txtMT_DID,'FontSize',fontSizeUsed+1);
set(f_handle,'CloseRequestFcn',{@mt_figure_close,h,f_handle});
set(f_handle,'KeyPressFcn',{@mt_figure_keypress,h,f_handle});
set(f_handle,'UserData', figureUserData);
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function mt_figure_keypress(obj, eventdata, h, f_handle)
% local function to handle KeyPress events on figure.
% Is used to (P)ause plot, (Z)oom in/out, (D)efault display,
% (A)ccelerometer only, rate (G)yro only, (M)agnetometer only
%
% envoked when a key is pressed when figure is in focus
in_key=lower(get(f_handle,'CurrentCharacter'));
tmp = get(f_handle,'UserData');
switch in_key
case ']' % view next connected sensor...
pause(0.2)
figure(f_handle);
tmp = get(f_handle,'UserData');
tmp(1) = 1;% indicate that next MT9 should be plotted
set(f_handle,'UserData', tmp);
case 'p' % pause to view graph
disp('Paused, press any key to continue...')
pause(0.2);% introduce a slight break, because otherwise 1 keystroke is recorded as multiple
figure(f_handle);% raise figure to foreground
pause; % wait for next key stroke
case 'z' % toggle zoom mode
pause(0.2)
figure(f_handle)
if tmp(2) == 0, % check zoom level
set(f_handle,'UserData',[tmp(1) 1 tmp(3)]); % toggle to next zoom mode
else
set(f_handle,'UserData',[tmp(1) 0 tmp(3)]); % toggle to default zoom mode
end
case 'r' % reset orientation
disp('Resetting heading orientation (boresight)...')
pause(0.2);% introduce a slight break, because otherwise 1 keystroke is recorded as multiple
figure(f_handle);% raise figure to foreground
XM_ResetOrientation(h,0,0); % default reset type 0 = heading, do not save to MTS = 0 (second parameter)
case 'a'
disp('Switching to display only 3D accelerometer data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 1]); % set to accelerometer mode
case 'g'
disp('Switching to display only 3D rate gyroscope data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 2]); % set to gyro mode
case 'm'
disp('Switching to display only 3D magnetometer data stream...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 3]); % set to mag mode
case 'd'
disp('Switching the default display mode, all 9 data streams...')
pause(0.2)
figure(f_handle)
set(f_handle,'UserData',[tmp(1:2) 0]); % set to default mode
case 'q'
disp('Quitting demo XM_DisplayRealtimeData...')
pause(0.2)
figure(f_handle)
close(f_handle)
otherwise
disp('Unknown command option....displaying help data')
disp(' ')
eval('help XM_DisplayRealtimeData')
end
if ishandle(f_handle) % needed to check if figure exists (using Q to quit)
if tmp(3)>3,% If in Orientation output mode, IGNORE any change in PlotMode!!!
tmp_new = get(f_handle,'UserData');
set(f_handle,'UserData',[tmp_new(1:2) tmp(3)]);
end
% reset CurrentCharater to no value...
set(f_handle,'CurrentCharacter',' ');
end
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function YLims = mt_plot_data(d, last_d, t, CurrentFigureUserData, p_handle)
% local function to plot the data using "low-level" set fucntions for smooth plotting
switch CurrentFigureUserData(3) % check plot type
case 0 %default
if CurrentFigureUserData(2), % check if zoomed
band = [0.8 40 0.1]; % define zoom range
YLims = [min(d(1,1:3))-band(1) max(d(1,1:3))+band(1); min(d(1,4:6)./pi.*180)-band(2) max(d(1,4:6)./pi.*180)+band(2);...
min(d(1,7:9))-band(3) max(d(1,7:9))+band(3)];
else % default values of zoom (full range of MT9)
YLims = [-22 22; -1200 1200; -1.8 1.8];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(4),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(7),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','r','LineWidth',2)
% convert the rate of turn data to deg/s instead of rad/s
set(p_handle(2),'XData',t,'YData',([last_d(1,4) d(:,4)'])./pi.*180,'Color','b','LineWidth',2)
set(p_handle(5),'XData',t,'YData',([last_d(1,5) d(:,5)'])./pi.*180,'Color','g','LineWidth',2)
set(p_handle(8),'XData',t,'YData',([last_d(1,6) d(:,6)'])./pi.*180,'Color','r','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,7) d(:,7)'],'Color','b','LineWidth',2)
set(p_handle(6),'XData',t,'YData',[last_d(1,8) d(:,8)'],'Color','g','LineWidth',2)
set(p_handle(9),'XData',t,'YData',[last_d(1,9) d(:,9)'],'Color','r','LineWidth',2)
case 1 % Only accelerometer
if CurrentFigureUserData(2), % check if zoomed
band = 0.8; % define zoom range (in m/s2)
YLims = [min(d(:,1))-band max(d(:,1))+band; min(d(:,2))-band max(d(:,2))+band; ...
min(d(:,3))-band max(d(:,3))+band];
else % default values of zoom (full range of MT9)
YLims = [-25 25; -25 25; -25 25];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',([last_d(1,3) d(:,3)']),'Color','r','LineWidth',2)
case 2 % Only rate gyros
if CurrentFigureUserData(2), % check if zoomed
band = 40; % define zoom range (in deg/s)
YLims = [min(d(:,4)./pi.*180)-band max(d(:,4)./pi.*180)+band;...
min(d(:,5)./pi.*180)-band max(d(:,5)./pi.*180)+band; min(d(:,6)./pi.*180)-band max(d(:,6)./pi.*180)+band];
else % default values of zoom (full range of MT9)
YLims = [-1200 1200; -1200 1200; -1200 1200];
end
% plot the data
set(p_handle(1),'XData',t,'YData',([last_d(1,4) d(:,4)'])./pi.*180,'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',([last_d(1,5) d(:,5)'])./pi.*180,'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',([last_d(1,6) d(:,6)'])./pi.*180,'Color','r','LineWidth',2)
case 3 % Only magnetometers
if CurrentFigureUserData(2), % check if zoomed
band = 0.1; % define zoom range (in a.u.)
YLims = [min(d(:,7))-band max(d(:,7))+band; min(d(:,8))-band max(d(:,8))+band; min(d(:,9))-band max(d(:,9))+band];
else % default values of zoom (full range of MT9)
YLims = [-2.5 2.5; -2.5 2.5; -2.5 2.5];
end
% plot the data
set(p_handle(1),'XData',t,'YData',[last_d(1,7) d(:,7)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,8) d(:,8)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,9) d(:,9)'],'Color','r','LineWidth',2)
case 4 % Euler angles
if CurrentFigureUserData(2), % zoom UserData
band = 6; % define zoom range (in degrees)
YLims = [min(d(:,1))-band max(d(:,1))+band; min(d(:,2))-band max(d(:,2))+band; ...
min(d(:,3))-band max(d(:,3))+band];
else % default values of zoom (full range of Euler angles)
YLims = [-180 180;-90 90;-180 180];
end
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','b','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','g','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','r','LineWidth',2)
case 5 % Quaternions
if CurrentFigureUserData(2), % zoom UserData
band = 0.14; % define zoom range
YLims = [min(d(1,1))-band max(d(1,1))+band; min(d(1,2))-band max(d(1,2))+band;...
min(d(1,3))-band max(d(1,3))+band; min(d(1,4))-band max(d(1,4))+band]; % not so useful for quaternions...
else % default values of zoom (full range of Euler angles)
YLims = [-1 1; -1 1; -1 1; -1 1;];
end
set(p_handle(1),'XData',t,'YData',[last_d(1,1) d(:,1)'],'Color','k','LineWidth',2)
set(p_handle(2),'XData',t,'YData',[last_d(1,2) d(:,2)'],'Color','b','LineWidth',2)
set(p_handle(3),'XData',t,'YData',[last_d(1,3) d(:,3)'],'Color','g','LineWidth',2)
set(p_handle(4),'XData',t,'YData',[last_d(1,4) d(:,4)'],'Color','r','LineWidth',2)
case 6 % Matrix
YLims = [0 4]; % not used
% only display latest data available in DCM orientation matrix mode
set(p_handle(1),'cdata',[d(end,1:3); d(end,4:6); d(end,7:9)]');
end % switch
% flush the graphics to screen
drawnow
% -------------------------------------------------------------------------
% end of function
%% -------------------------------------------------------------------------
function mt_figure_close(obj, eventdata, h, f_handle)
% local function to properly release MTObj when the user kills the figure window.
% release MTObj
XM_StopProcess(h)
delete(h);
clear h;
% kill figure window as requested
delete(f_handle)
% -------------------------------------------------------------------------
% end of function
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildTensorStyleActRep.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/buildTensorStyleActRep.m
| 6,442 |
utf_8
|
ee9af2e2baeb0930757e0db140fa496b
|
function [Tensor,mots,skels]=buildTensorStyleActRep(p,varargin)
% T=buildTensorStyleActRep(dir,[maxRep[,dataRep[,Styles]]]);
% example:
% T=buildTensorStyleActRep('R:\HDM05\HDM05_cut_amc',3);
switch nargin
case 1
maxRep =3;
dataRep='Quat';
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 2
maxRep =varargin{1};
dataRep='Quat';
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 3
maxRep =varargin{1};
dataRep=varargin{2};
Styles ={'walk4StepsRstart', ...
'walkLeftCircle4StepsRstart', ...
'walkRightCircle4StepsRstart'};
case 4
maxRep =varargin{1};
dataRep=varargin{2};
Styles =varargin{3};
otherwise
error('Wrong number of Args');
end
Tensor.DataRep=dataRep;
% define size of representation:
switch dataRep
case 'Quat'
dimDataRep=4;
case 'Position'
dimDataRep=3;
case 'ExpMap'
dimDataRep=3;
case 'Acc'
dimDataRep=3;
otherwise
error('buildTensorStyleActRep_jt: Wrong Date specified in var: dataRep');
end
% Check if Backslash is included in path and append extension:
if(p(end)~=filesep)
p=[p filesep];
end
ext='*.amc';
% Check if there is a directory for every style:
numStyles=size(Styles,2);
for s=1:numStyles
if(~exist([p Styles{1,s}],'dir'))
error(['buildTensorStyleActRep_jt: Dir for Style ' Styles{1,s} ' does not exist!']);
end
end
Tensor.styles=Styles;
% Get Lists of files:
for s=1:numStyles
listofFiles{s}=dir([p Styles{1,s} filesep ext]);
end
%% Collect Information about the given files/styles/classes:
% Known Actors: bd,bk,mm,dg,tr
actors{1}='HDM_bd';
actors{2}='HDM_bk';
actors{3}='HDM_dg';
actors{4}='HDM_mm';
actors{5}='HDM_tr';
numActors=size(actors,2);
for s=1:numStyles
%Count repetitions of each actor
for a=1:numActors
reps(a,s)=countActor(listofFiles{s},actors{a});
end
end
% Motions to fit by DTW
skelfile=fullfile(p, Styles{1,1}, [actors{1} '.asf']);
motfile =fullfile(p, Styles{1,1}, listofFiles{1}(1).name);
[fitskel,fitmot]=readMocap(skelfile,motfile);
fprintf('Reference Motion: ');
fprintf([listofFiles{1}(1).name '\n'])
fitmot=changeFrameRate(fitskel,fitmot,30);
Tensor.numTechnicalModes = 3;
Tensor.dimTechnicalModes = [dimDataRep fitmot.nframes fitmot.njoints];
Tensor.dimNaturalModes = [numStyles numActors maxRep];
Tensor.dimNaturalModes = Tensor.dimNaturalModes(Tensor.dimNaturalModes~=1);
Tensor.numNaturalModes = length(Tensor.dimNaturalModes);
mots = cell(numStyles,numActors,maxRep);
skels = cell(numStyles,numActors,maxRep);
Tensor.data=zeros([Tensor.dimTechnicalModes Tensor.dimNaturalModes]);
for s=1:numStyles
file=1;
for a=1:numActors
for r=1:max(maxRep,reps(a,s))
if (r<=maxRep)
fprintf('Fitting motion %i%i%i - ',s,a,r);
skelfile = fullfile(p, Styles{1,s}, [actors{a} '.asf']);
motfile = fullfile(p, Styles{1,s}, listofFiles{s}(file).name);
Tensor.motions{s,a,r} = motfile;
Tensor.skeletons{s,a,r} = skelfile;
[skel,mot] = readMocap(skelfile,motfile);
mot = changeFrameRate(skel,mot,30);
mot = fitMotion(skel,mot);
[D,w] = pointCloudDTW(fitmot,mot,'pos',1:31,1);
mot = warpMotion(w,skel,mot);
Tensor.rootdata(:,:,s,a,r) = mot.rootTranslation;
for joint=1:mot.njoints
% Tensor.joints{joint,s,a,r}=mot.jointNames{joint};
switch dataRep
case 'Quat'
if(~isempty(mot.rotationQuat{joint}))
% align all quaternions
mot.rotationQuat{joint}(:,dot(mot.rotationQuat{joint},fitmot.rotationQuat{joint})<0)...
=-mot.rotationQuat{joint}(:,dot(mot.rotationQuat{joint},fitmot.rotationQuat{joint})<0);
Tensor.data(:,:,joint,s,a,r)=mot.rotationQuat{joint};
else
Tensor.data(1,:,joint,s,a,r) =1;
Tensor.data(2:4,:,joint,s,a,r)=0;
end
case 'Position'
if(~isempty(mot.jointTrajectories{joint}))
Tensor.data(:,:,joint,s,a,r)=mot.jointTrajectories{joint};
else
Tensor.data(:,:,joint,s,a,r)=0;
end
case 'ExpMap'
if(~isempty(mot.rotationQuat{joint}))
Tensor.data(:,:,joint,s,a,r)=quatlog(mot.rotationQuat{joint});
else
Tensor.data(:,:,joint,s,a,r)=0;
end
case 'Acc'
mot = addAccToMot(mot);
if(~isempty(mot.jointAccelerations{joint}))
Tensor.data(:,:,joint,s,a,r)=mot.jointAccelerations{joint};
else
Tensor.data(:,:,joint,s,a,r)=0;
end
otherwise
error('buildTensorStyleActRep_jt: Wrong Type specified in var: dataRep\n');
end
end
[mots{s,a,r},skels{s,a,r}]=deal(mot,skel);
end
if (r<reps(a,s)||r==max(maxRep,reps(a,s)))
file=file+1;
end
end
end
end
Tensor.data=squeeze(Tensor.data);
Tensor.rootdata=squeeze(Tensor.rootdata);
Tensor=HOSVDv2(Tensor);
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
compareAngles.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/compareAngles.m
| 1,720 |
utf_8
|
af58585e18a5b41ed729fc6a30ae9b47
|
function [res] = compareAngles(mot1,mot2,varargin)
mot1=addBonesToMot(mot1);
mot2=addBonesToMot(mot2);
endEffectors = [6,11,17,23,24,30,31];
regardedJoints = setxor((1:31),endEffectors);
weights = ones(1,length(regardedJoints));
switch nargin
case 2
case 3
if ~isempty(intersect(varargin{1},endEffectors))
fprintf('End effectors selected - ignoring end effectors...\n');
end
regardedJoints = intersect(varargin{1},regardedJoints);
weights = ones(1,length(regardedJoints));
case 4
if length(varargin{1})~=length(varargin{2})
error('Number of selected joints does not equal number of weights!');
end
if ~isempty(intersect(varargin{1},endEffectors))
fprintf('End effectors selected - ignoring end effectors...\n');
end
[regardedJoints,pos1] = intersect(varargin{1},regardedJoints);
weights = varargin{2}(pos1);
otherwise
error('Wrong number og argins!');
end
fprintf('Joints: \t');
fprintf('%.3i ',regardedJoints);
fprintf('\nWeights: \t');
fprintf('%.1f ',weights);
fprintf('\n');
counter=0;
for i=regardedJoints
counter=counter+1;
res(counter,:)=weights(counter)*computeAngleDiff(mot1,mot2,i);
end
function diff=computeAngleDiff(mot1,mot2,joint)
switch joint
case 3 %lknee
z = arrayfun(@(x)intersect(x.f1, x.f2),mot1.bones{:,5})
end
angle1=real(acosd(dot(-mot1.bones{2,4},mot1.bones{3,4})));
angle2=real(acosd(dot(-mot2.bones{2,4},mot2.bones{3,4})));
dot(cross(-mot1.bones{2,4},mot1.bones{3,4}),cross(-mot2.bones{2,4},mot2.bones{3,4}))
diff=abs(angle1-angle2);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
diff5point2.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/diff5point2.m
| 2,637 |
utf_8
|
d17ccf79ec510a81dd4d2db6d2064f3a
|
% function newStream = diff5point(stream,samplingRate,varargin)
% 5-point derivation
% author: Jochen Tautges ([email protected])
function newStream = diff5point2(stream,samplingRate,varargin)
% defaults
options.padding = true;
options.derivative = 1;
options.side = 'center';
nrOfFrames = size(stream,2);
if nargin>2
options = mergeOptions(varargin{1},options);
end
if options.padding
newStream = zeros(size(stream));
else
newStream = zeros(size(stream,1),nrOfFrames-4);
end
prepadding = [];
postpadding = [];
switch options.side
case 'left'
left = 4;
right = 0;
if options.padding
prepadding = interp1(1:nrOfFrames,stream',-3:0,'linear','extrap')';
postpadding = [];
end
switch options.derivative
case 1
weights = [3 -16 36 -48 25];
divisor = 12/samplingRate;
case 2
weights = -[-11 56 -114 104 -35];
divisor = 12/samplingRate^2;
otherwise
error('Not yet implemented!');
end
case 'right'
left = 0;
right = 4;
if options.padding
prepadding = [];
postpadding = interp1(1:nrOfFrames,stream',nrOfFrames+1:nrOfFrames+4,'linear','extrap')';
end
switch options.derivative
case 1
weights = [-25 48 -36 16 -3];
divisor = 12/samplingRate;
case 2
weights = [35 -104 114 -56 11];
divisor = 12/samplingRate^2;
otherwise
error('Not yet implemented!');
end
case 'center'
left = 2;
right = 2;
if options.padding
prepadding = interp1(1:nrOfFrames,stream',-1:0,'linear','extrap')';
postpadding = interp1(1:nrOfFrames,stream',nrOfFrames+1:nrOfFrames+2,'linear','extrap')';
end
switch options.derivative
case 1
weights = [1 -8 0 8 -1];
divisor = 12/samplingRate;
case 2
weights = [-1 16 -30 16 -1];
divisor = 12/samplingRate^2;
otherwise
error('Not yet implemented!');
end
end
if size(stream,1)==1
prepadding = prepadding';
postpadding = postpadding';
end
stream = [prepadding stream postpadding];
for i = left+1:size(stream,2)-right
newStream(:,i-left) = stream(:,i-left:i+right) * weights';
end
newStream = newStream / divisor;
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
buildTensor.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/buildTensor.m
| 20,295 |
utf_8
|
dcc2bdb7af8b33478e4603d910c363b5
|
% function buildTensor
% builds a motion tensor with specified form and order of modes
%
% Use upper cases for Technical Modes, lower cases for Natural Modes!
% Valid characters for Technical Modes:
% I: (Q)uaternions / (E)uler angles / Exponential (M)aps / (P)ositions /
% (V)elocities / (A)ccelerations; (necessary)
% II: (F)rames (necessary)
% III: (J)oints (necessary)
% Valid characters for Natural Modes:
% i: (a)ctors (necessary)
% ii: (s)tyles (necessary)
% iii: (r)epetitions (necessary)
% iv: (m)irrored motions (optional)
%
% [Tensor,skels,mots] = buildTensor(form)
% example: T = buildTensor('QFJsar');
% author: Jochen Tauges ([email protected])
function [T,skels,mots]=buildTensor(form)
home='R:\HDM05\';
if (isempty(strfind(form,'J')))...
|| (isempty(strfind(form,'F')))...
|| (isempty(strfind(form,'a')))...
|| (isempty(strfind(form,'s')))...
|| (isempty(strfind(form,'r')))
error('Invalid input!');
end
T.DataRep = [];
T.numTechnicalModes = 0;
T.numNaturalModes = 0;
dimDataRep = 3;
T.dimTechnicalModes = [];
T.dimNaturalModes = [];
nrOfActors = 0;
nrOfStyles = 0;
nrOfRepetitions = 0;
mirroring = false;
TechOrder = zeros(1,sum(int8(form)<91));
NatOrder = zeros(1,sum(int8(form)>97));
for i=1:length(form)
if length(strfind(form,form(i)))>1
error('Invalid input!');
end
switch form(i)
case 'Q'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'QUAT';
T.DataRep = 'quat';
dimDataRep = 4;
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'E'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'EULER';
T.DataRep = 'euler';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'M'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'EXPMAP';
T.DataRep = 'expmap';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'P'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'POS';
T.DataRep = 'pos';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'V'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'VEL';
T.DataRep = 'vel';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'A'
if (T.numNaturalModes>0 || ~isempty(T.DataRep))
error('Invalid input!');
else
T.form{i} = 'ACC';
T.DataRep = 'acc';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 1;
end
case 'F'
if (T.numNaturalModes>0)
error('Invalid input!');
else
T.form{i} = 'FRAMES';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 2;
end
case 'J'
if (T.numNaturalModes>0)
error('Invalid input!');
else
T.form{i} = 'JOINTS';
T.numTechnicalModes = T.numTechnicalModes+1;
TechOrder(i) = 3;
end
jointModeID = i;
case 's'
T.form{i} = 'styles';
NatOrder(i-T.numTechnicalModes) = 4;
T.numNaturalModes = T.numNaturalModes+1;
case 'a'
T.form{i} = 'actors';
NatOrder(i-T.numTechnicalModes) = 5;
T.numNaturalModes = T.numNaturalModes+1;
case 'r'
T.form{i} = 'repetitions';
NatOrder(i-T.numTechnicalModes) = 6;
T.numNaturalModes = T.numNaturalModes+1;
while nrOfRepetitions<=0
nrOfRepetitions = input('How many repetitions of each motion do you want to store? ');
end
fprintf('Size of repetition mode: %i\n',nrOfRepetitions);
case 'm'
T.form{i} = 'mirrors';
mirroring = true;
NatOrder(i-T.numTechnicalModes) = 7;
T.numNaturalModes = T.numNaturalModes+1;
otherwise
help buildTensor;
error('Invalid character: ''%c''!',form(i));
end
end
extension='*.amc';
if exist(home,'dir');
AMCfolder=home;
else
AMCfolder='';
end
% Hack for tensor construction EG08 ;-)
AMCfolder='/data/HDM/HDM05_EG08/HDM05_EG08_cut_amc_training/';
%reading the .asf-files:
FilterIndex=1;
while (FilterIndex~=0 || nrOfActors==0)
fprintf('\nSpecify asf-file(s)! ');
[ASFfile,ASFPathName,FilterIndex] = uigetfile(fullfile(AMCfolder,'*.asf'),'MultiSelect','on');
if iscell(ASFfile)
for j=1:length(ASFfile)
fprintf('%s ',ASFfile{j});
nrOfActors = nrOfActors+1;
actors{nrOfActors} = ASFfile{j};
actorsFullFile{nrOfActors} = fullfile(ASFPathName, ASFfile{j});
end
else
if FilterIndex~=0
fprintf('%s ',ASFfile);
nrOfActors = nrOfActors+1;
actors{nrOfActors} = ASFfile;
actorsFullFile{nrOfActors} = fullfile(ASFPathName, ASFfile);
elseif nrOfActors>0
fprintf('\nSize of actors mode: %i\n',nrOfActors);
end
end
end
%reading the .amc files
%file names have to match the .asf file names!
while (ischar(AMCfolder) || nrOfStyles==0)
dialog_title=['Specify folder of amc-files for style ' num2str(nrOfStyles+1) '!'];
fprintf('\n%s ',dialog_title);
if AMCfolder==0
AMCfolder='';
end
AMCfolder = uigetdir(AMCfolder,dialog_title);
if (AMCfolder~=0)
nrOfStyles = nrOfStyles+1;
for a=1:nrOfActors
actorString{a} = actors{a}(1:end-4);
listOfAMCFiles_tmp = dir(fullfile(AMCfolder,['*',actorString{a},'*',extension]));
listOfAMCFiles{nrOfStyles,a} = listOfAMCFiles_tmp;
reps(a,nrOfStyles) = size(listOfAMCFiles{nrOfStyles,a},1);
end
AMCpaths{nrOfStyles} = AMCfolder;
T.styles{nrOfStyles} = fliplr(strtok(fliplr(AMCfolder),filesep));
else
fprintf('\nSize of style mode: %i\n',nrOfStyles);
end
end
refMotFileName = fullfile(AMCpaths{1},listOfAMCFiles{1,1}(1).name);
fprintf('\nSelect reference motion for DTW (Cancel for default)!\n ');
[refFileName,refPathName,FilterIndex] = uigetfile(fullfile(AMCpaths{1},'*.amc'));
if FilterIndex~=0
refMotFileName=fullfile(refPathName,refFileName);
end
refSkelFileName=[];
for a=1:nrOfActors
if size(strfind(refMotFileName,actorString{a}))>0
refSkelFileName=actorsFullFile{a};
break;
end
end
if isempty(refSkelFileName)
[refFileName,refSkelPathName,FilterIndex] = uigetfile(fullfile(ASFPathName,'*.asf'), 'Choose ASF file for reference motion!');
if FilterIndex~=0
refSkelFileName=fullfile(refSkelPathName,refFileName);
end
end
[fitskel,fitmot] = readMocap(refSkelFileName,refMotFileName);
%%% Vorsicht: readMocap liest samplingRate nicht korrekt ein! default=120
% fitmot.samplingRate = 30;
% fitmot.frametime = 1/30;
T.samplingRate = input('\nSpecify frame rate for all motions: ');
fitmot = changeFrameRate(fitskel,fitmot,T.samplingRate);
fitmot = fitMotion(fitskel,fitmot);
T.DTW.refMot = fliplr(strtok(fliplr(fitmot.filename),filesep));
for i=1:length(form)
switch form(i)
case {'Q','E','M','P','V','A'}
T.dimTechnicalModes = [T.dimTechnicalModes dimDataRep];
case 'F'
T.dimTechnicalModes = [T.dimTechnicalModes fitmot.nframes];
case 'J'
T.dimTechnicalModes = [T.dimTechnicalModes fitmot.njoints];
case 'a'
T.dimNaturalModes = [T.dimNaturalModes nrOfActors];
case 's'
T.dimNaturalModes = [T.dimNaturalModes nrOfStyles];
case 'r'
T.dimNaturalModes = [T.dimNaturalModes nrOfRepetitions];
case 'm'
T.dimNaturalModes = [T.dimNaturalModes 2];
end
end
T.form = T.form([(T.dimTechnicalModes>1) (T.dimNaturalModes>1)]);
T.dimTechnicalModes = T.dimTechnicalModes(T.dimTechnicalModes>1);
T.dimNaturalModes = T.dimNaturalModes(T.dimNaturalModes>1);
T.numNaturalModes = length(T.dimNaturalModes);
T.numTechnicalModes = length(T.dimTechnicalModes);
if mirroring
mots = cell(nrOfStyles,nrOfActors,nrOfRepetitions,2);
skels = cell(nrOfStyles,nrOfActors,nrOfRepetitions,2);
T.motions = cell(nrOfStyles,nrOfActors,nrOfRepetitions,2);
T.skeletons = cell(nrOfStyles,nrOfActors,nrOfRepetitions,2);
else
mots = cell(nrOfStyles,nrOfActors,nrOfRepetitions);
skels = cell(nrOfStyles,nrOfActors,nrOfRepetitions);
T.motions = cell(nrOfStyles,nrOfActors,nrOfRepetitions);
T.skeletons = cell(nrOfStyles,nrOfActors,nrOfRepetitions);
end
motfile_tmp = '';
T.numMissingMots = 0;
T.DTW.warpingCosts = 0;
for s=1:nrOfStyles
for a=1:nrOfActors
file=1;
for r=1:max(nrOfRepetitions,reps(a,s))
if (r<=nrOfRepetitions)
skelfile = actorsFullFile{a};
motfile = fullfile(AMCpaths{s},listOfAMCFiles{s,a}(file).name);
T.motions{s,a,r,1} = motfile;
T.skeletons{s,a,r,1} = skelfile;
if strcmp(motfile,refMotFileName)
if strcmp(motfile,motfile_tmp)
fprintf('\nMotion %i%i%i is not existent and set equal to previous motion.\n',s,a,r);
else
fprintf('\nMotion %i%i%i is the reference motion of the DTW and does not have to be aligned again.\n',s,a,r);
skel = fitskel;
mot = fitmot;
if ~isfield(T.DTW,'refMotID')
T.DTW.refMotID=[s,a,r,1];
end
end
T.DTW.warpingPaths{s,a,r,1} = repmat((mot.nframes:-1:1)',1,2);
elseif strcmp(motfile,motfile_tmp)
fprintf('\nMotion %i%i%i is not existent and set equal to previous motion.\n',s,a,r);
T.DTW.warpingPaths{s,a,r,1} = w;
T.DTW.warpingCosts = [T.DTW.warpingCosts D(fitmot.nframes,mot.nframes)/fitmot.nframes];
T.numMissingMots = T.numMissingMots+1;
else
fprintf('\nFitting motion %i%i%i - ',s,a,r);
[skel,mot] = readMocap(skelfile,motfile);
mot = changeFrameRate(skel,mot,T.samplingRate);
mot = fitMotion(skel,mot);
[D,w] = pointCloudDTW_pos(fitmot,mot,2);
T.DTW.warpingPaths{s,a,r,1} = w;
T.DTW.warpingCosts = [T.DTW.warpingCosts D(fitmot.nframes,mot.nframes)/fitmot.nframes];
mot = warpMotion(w,skel,mot);
motfile_tmp = motfile;
end
T.rootdata(:,:,s,a,r,1) = mot.rootTranslation;
for joint=1:mot.njoints
switch lower(T.DataRep)
case 'quat'
if(~isempty(mot.rotationQuat{joint}) && ~isempty(fitmot.rotationQuat{joint}))
% align all quaternions
mot.rotationQuat{joint}(:,dot(mot.rotationQuat{joint},fitmot.rotationQuat{joint})<0)...
= -mot.rotationQuat{joint}(:,dot(mot.rotationQuat{joint},fitmot.rotationQuat{joint})<0);
T.data(:,:,joint,s,a,r,1) = mot.rotationQuat{joint};
else
T.data(1,:,joint,s,a,r) = 1;
T.data(2:4,:,joint,s,a,r) = 0;
end
case 'euler'
[xxx,mot]=convert2euler(skel,mot);
if(~isempty(mot.rotationEuler{joint}))
T.data(:,:,joint,s,a,r,1) = mot.rotationEuler{joint};
else
T.data(:,:,joint,s,a,r,1) = 0;
end
case 'expmap'
if(~isempty(mot.rotationQuat{joint}))
T.data(:,:,joint,s,a,r,1) = quatlog(mot.rotationQuat{joint});
else
T.data(:,:,joint,s,a,r,1) = 0;
end
case 'pos'
if(~isempty(mot.jointTrajectories{joint}))
T.data(:,:,joint,s,a,r,1) = mot.jointTrajectories{joint};
else
T.data(:,:,joint,s,a,r,1) = 0;
end
case 'vel'
mot = addVelToMot(mot);
if(~isempty(mot.jointVelocities{joint}))
T.data(:,:,joint,s,a,r,1) = mot.jointVelocities{joint};
else
T.data(:,:,joint,s,a,r,1) = 0;
end
case 'acc'
mot = addAccToMot(mot);
if(~isempty(mot.jointAccelerations{joint}))
T.data(:,:,joint,s,a,r,1) = mot.jointAccelerations{joint};
else
T.data(:,:,joint,s,a,r,1) = 0;
end
end % (switch lower(T.DataRep))
end % (for joint=1:motM.njoints)
[skels{s,a,r,1},mots{s,a,r,1}] = deal(skel,mot);
if mirroring
[skelM,motM] = mirrorMot(skel,mot);
motM = fitMotion(skelM,motM);
[skels{s,a,r,2},mots{s,a,r,2}] = deal(skelM,motM);
T.motions{s,a,r,2} = [motfile '.mirrored'];
T.skeletons{s,a,r,2} = [skelfile '.mirrored'];
T.DTW.warpingPaths{s,a,r,2} = T.DTW.warpingPaths{s,a,r,1};
T.rootdata(:,:,s,a,r,2) = motM.rootTranslation;
for joint=1:motM.njoints
switch lower(T.DataRep)
case 'quat'
if(~isempty(motM.rotationQuat{joint}))
% aligning not necessary anymore
T.data(:,:,joint,s,a,r,2) = motM.rotationQuat{joint};
else
T.data(1,:,joint,s,a,r,2) = 1;
T.data(2:4,:,joint,s,a,r,2) = 0;
end
case 'euler'
[xxx,motM]=convert2euler(skelM,motM);
if(~isempty(motM.rotationEuler{joint}))
T.data(:,:,joint,s,a,r,2) = motM.rotationEuler{joint};
else
T.data(:,:,joint,s,a,r,2) = 0;
end
case 'expmap'
if(~isempty(motM.rotationQuat{joint}))
T.data(:,:,joint,s,a,r,2) = quatlog(motM.rotationQuat{joint});
else
T.data(:,:,joint,s,a,r,2) = 0;
end
case 'pos'
if(~isempty(motM.jointTrajectories{joint}))
T.data(:,:,joint,s,a,r,2) = motM.jointTrajectories{joint};
else
T.data(:,:,joint,s,a,r,2) = 0;
end
case 'vel'
motM = addVelToMot(motM);
if(~isempty(motM.jointVelocities{joint}))
T.data(:,:,joint,s,a,r,2) = motM.jointVelocities{joint};
else
T.data(:,:,joint,s,a,r,2) = 0;
end
case 'acc'
motM = addAccToMot(motM);
if(~isempty(motM.jointAccelerations{joint}))
T.data(:,:,joint,s,a,r,2) = motM.jointAccelerations{joint};
else
T.data(:,:,joint,s,a,r,2) = 0;
end
end % (switch lower(T.DataRep))
end % (for joint=1:motM.njoints)
end % (if mirroring)
end % (if (r<=nrOfRepetitions))
if (r<reps(a,s)||r==max(nrOfRepetitions,reps(a,s)))
file=file+1;
end
end
end
end
dataOrder = [TechOrder NatOrder];
rootdataOrder = dataOrder(dataOrder~=jointModeID);
rootdataOrder(rootdataOrder>jointModeID)=...
rootdataOrder(rootdataOrder>jointModeID)-1;
order = NatOrder-T.numTechnicalModes;
T.data = squeeze(permute(T.data,dataOrder));
T.rootdata = squeeze(permute(T.rootdata,rootdataOrder));
T.motions = squeeze(permute(T.motions,order));
T.skeletons = squeeze(permute(T.skeletons,order));
T.DTW.warpingPaths = squeeze(permute(T.DTW.warpingPaths,order));
skels = squeeze(permute(skels,order));
mots = squeeze(permute(mots,order));
if isfield(T.DTW,'refMotID')
T.DTW.refMotID = T.DTW.refMotID(order);
T.DTW.refMotID = T.DTW.refMotID(T.dimNaturalModes>1);
end
T = HOSVDv2(T);
end
function num=countActor(files,actor)
LoFN=[files(:).name];
tmp=size(strfind(LoFN,actor));
num=tmp(2);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
pointCloudDTW.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/pointCloudDTW.m
| 3,102 |
utf_8
|
83f6115f6c0f4180cbd4fdd61dcf3716
|
% function [D,w,d]=pointCloudDTW(motRef,motToWarp[,traj,joints,rho])
%
% Output:
% D = GDM (Global Distance Matrix)
% w = warping path
% d = LDM (Local Distance Matrix)
%
% Required Input:
% motref: reference motion
% motToWarp: motion which is to warp
%
% Optional input:
% traj: 'jointTrajectories' (default), 'jointVelocities' or
% 'jointAccelerations'
% joints: joints for which the trajectories are to be compared
% (for joint IDs see the skel structure, default: [1:31])
% rho: windowsize for point cloud distance (default: 1)
function [D,w,d]=pointCloudDTW(motRef,motToWarp,varargin)
% default values
trajField='jointTrajectories';
joints=(1:31);
rho=0;
switch nargin
case 3
trajField=varargin{1};
case 4
trajField=varargin{1};
joints=varargin{2};
case 5
trajField=varargin{1};
joints=varargin{2};
rho = varargin{3};
end
TrajectoriesRef = cell(length(joints),1);
TrajectoriesToWarp = cell(length(joints),1);
% if ~(isfield(motRef,'jointAccelerations'))
% motRef=addAccToMot(motRef);
% end
% if ~(isfield(motToWarp,'jointAccelerations'))
% motToWarp=addAccToMot(motToWarp);
% end
for i=1:length(joints)
switch trajField
case {'jointAccelerations','a','acc','A','Acc','accelerations','Accelerations'}
TrajectoriesRef{i} = motRef.jointAccelerations{joints(i)};
TrajectoriesToWarp{i} = motToWarp.jointAccelerations{joints(i)};
case {'jointVelocities','v','vel','V','Vel','velocities','Velocities'}
TrajectoriesRef{i} = motRef.jointVelocities{joints(i)};
TrajectoriesToWarp{i} = motToWarp.jointVelocities{joints(i)};
case {'jointTrajectories','p','pos','P','Pos','positions','Positions'}
TrajectoriesRef{i} = motRef.jointTrajectories{joints(i)};
TrajectoriesToWarp{i} = motToWarp.jointTrajectories{joints(i)};
otherwise
fprintf('\nNo valid trajectory specification: %s\n\n', trajField);
[D,w,d]=deal([],[],[]);
help pointCloudDTW_jt
return
end
end
N = size(TrajectoriesRef{1},2);
M = size(TrajectoriesToWarp{1},2);
% Calculation of the LDM:
d = distMatrix_pointCloudDistance_jt(TrajectoriesRef,TrajectoriesToWarp,rho);
% Calculation of the GDM:
D = inf(size(d));
D(1,:) = cumsum(d(1,:));
D(:,1) = cumsum(d(:,1));
for n=2:N
for m=2:M
D(n,m) = d(n,m)+min([D(n-1,m-1),D(n-1,m),D(n,m-1)]);
end
end
% Search for the optimal path on the GDM:
n=N;
m=M;
w=[];
w(1,:)=[N,M];
while (n~=1 && m~=1)
[values,number]=min([D(n-1,m-1),D(n-1,m),D(n,m-1)]);
switch number
case 2
n=n-1;
case 3
m=m-1;
case 1
n=n-1;
m=m-1;
end
w=[w;[n,m]];
end
if (n==1 && m>1)
w=[w;[ones(m-1,1),(m-1:-1:1)']];
elseif (m==1 && n>1)
w=[w;[(n-1:-1:1)',ones(n-1,1)]];
end
% plotDTWpath(d,w);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
resampleMot2.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/resampleMot2.m
| 3,220 |
utf_8
|
d8f75da0022de6825c0010d62c4eec37
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% function resampleMot2
% for doc see "resampleMot"
% resampleMot2 uses arbitrary interpolation method instead of linear interpolation
% Note: resampleMot is faster, but resampleMot2 is nicer!
%
% author: Jochen Tautges ([email protected])
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function [resmot,newSamp,oldSamp] = resampleMot2(skel,mot,varargin)
rootTranslation = mot.rootTranslation;
mot.rootTranslation(:,:) = 0;
mot.jointTrajectories = C_forwardKinematicsQuat(skel,mot);
switch nargin
case 2
numberOfSamples = mot.nframes;
joints = 1:mot.njoints;
method = 'spline';
case 3
numberOfSamples = varargin{1};
joints = 1:mot.njoints;
method = 'spline';
case 4
numberOfSamples = varargin{1};
joints = varargin{2};
method = 'spline';
case 5
numberOfSamples = varargin{1};
joints = varargin{2};
method = varargin{3};
otherwise
help resampleMot;
error('Wrong number of argins!');
end
data = cell2mat(mot.jointTrajectories(joints));
dists = sqrt(sum(diff(data,1,2).^2));
delta = sum(dists)/(numberOfSamples-1);
dists = [0 cumsum(dists)];
oldSamp = 1:mot.nframes;
newSamp = zeros(1,numberOfSamples);
newSamp(1) = 1;
newSamp(end) = mot.nframes;
c=1;
i=2;
d = delta;
while i<=size(dists,2)
if dists(i)>delta
c=c+1;
r=dists(i)-dists(i-1);
p=delta-dists(i-1);
newSamp(c) = i-1+p/r;
delta = delta+d;
else
i=i+1;
end
end
dofs = getDOFsFromSkel(skel);
resmot = emptyMotion(mot);
resmot.nframes = numberOfSamples;
rotationQuat = cell2mat(mot.rotationQuat(mot.animated));
% rotationQuat = spline(oldSamp,rotationQuat,newSamp);
rotationQuat = interp1(oldSamp,rotationQuat',newSamp,method)';
resmot.rotationQuat = mat2cell(rotationQuat,dofs.quat,numberOfSamples);
resmot.rotationQuat = cellfun(@(x) normalizeColumns(x),resmot.rotationQuat,'UniformOutput',0);
resmot.rootTranslation = spline(oldSamp,rootTranslation,newSamp);
if isfield(mot,'jointAccelerations')
jointAccelerations = cell2mat(mot.jointAccelerations);
jointAccelerations = spline(oldSamp,jointAccelerations,newSamp);
resmot.jointAccelerations = mat2cell(jointAccelerations,dofs.pos,numberOfSamples);
end
if isfield(mot,'jointVelocities')
jointVelocities = cell2mat(mot.jointVelocities);
jointVelocities = spline(oldSamp,jointVelocities,newSamp);
resmot.jointVelocities = mat2cell(jointVelocities,dofs.pos,numberOfSamples);
end
resmot.jointTrajectories = C_forwardKinematicsQuat(skel,resmot);
resmot.boundingBox = computeBoundingBox(resmot);
resmot.rotationEuler = [];
resmot.documentation = 'resampled';
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
addDOFIDsToSkel.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/addDOFIDsToSkel.m
| 646 |
utf_8
|
11e095a26ec25bb14198480b779910f8
|
% function addDOFIDsToSkel
% adds struct "DOFIDs" to each node of denoted skel with indices
% identifying its respective (rotational) dofs (->efficiency)
% x -> 1, y -> 2, z -> 3
% example:
% skel = addDOFIDsToSkel(skel)
% author: Jochen Tautges, [email protected]
function skel = addDOFIDsToSkel(skel)
for i=1:skel.njoints
skel.nodes(i).DOFIDs = [];
for j=1:size(skel.nodes(i).DOF,1)
dof = lower(skel.nodes(i).DOF{j});
if strcmp(dof(1),'r');
skel.nodes(i).DOFIDs = [skel.nodes(i).DOFIDs ...
strfind(lower(skel.nodes(i).rotationOrder),dof(2))];
end
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructMotionFromWiiData.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/reconstructMotionFromWiiData.m
| 4,757 |
utf_8
|
bd78c7bd53046a5480a073842545f393
|
function res = reconstructMotionFromWiiData(Tensor,skel,wii_output,motOrig)
joint = input('Choose jointID for wii output: ');
% wii_output = wii_output(:,9:11)';
% wii_output(3,:) = -9.81 * wii_output(3,:);
% wii_output = wii_output(:,1:3)';
% wii_output(3,:) = -wii_output(3,:);
X0 = cell(1,Tensor.numNaturalModes);
x0 = [];
% X0{1}=[0 1 0];
% X0{2}=[1 0 0 0 0];
% X0{3}=[1 0 0];
for i=1:Tensor.numNaturalModes
X0{i} = ones(1,Tensor.dimNaturalModes(i))/Tensor.dimNaturalModes(i);
x0 = [x0 X0{i}];
end
% x0 = [1 0 0 1 0 0 0 0 1 0 0];
% [xx,refMot] = constructMotion(Tensor,X0,skel);
% refMot = addAccToMot(refMot);
% refdata = refMot.jointAccelerations{joint};
% % refdata=mean{joint};
% refdata(2,:) = refdata(2,:) + 9.81;
% % framerate
% oldRate = 50;
% newRate = 30;
% wii_output = changeSamplingRate(wii_output,oldRate,newRate);
% wii_output = filterTimeline(wii_output,2,'bin');
% wiiData = cutWiiData(wii_output,refdata);
% wiiData = resample(wii_output',floor(86/length(wii_output)*1000),1000)';
wiiData = wii_output;
% DTW
options = optimset( 'Display','iter',...
'MaxFunEvals',100000,...
'MaxIter',100,...
'TolFun',0.001,...
'PlotFcns', @optimplotx);
core = Tensor.core;
rootcore = Tensor.rootcore;
for i=1:Tensor.numTechnicalModes
core = modeNproduct(core,Tensor.factors{i},i);
end
for i=1:Tensor.numTechnicalModes-1
rootcore = modeNproduct(rootcore,Tensor.rootfactors{i},i);
end
optimStruct.motOrig = motOrig;
optimStruct.wiiData = wiiData;
optimStruct.joint = joint;
optimStruct.tensor = Tensor;
optimStruct.preparedCore = core;
optimStruct.preparedRootCore = rootcore;
optimStruct.skel = skel;
lb = -0.1 * ones(1,length(x0));
ub = 1.1 * ones(1,length(x0));
coeffs = lsqnonlin(@(x) objfunWii(x,optimStruct),x0,lb,ub,options);
res.coeffs = mat2cell(coeffs,1,Tensor.dimNaturalModes)';
for i=1:Tensor.numNaturalModes
core = modeNproduct(core,...
res.coeffs{i}*Tensor.factors{i+Tensor.numTechnicalModes},...
i+Tensor.numTechnicalModes);
end
for i=1:Tensor.numNaturalModes
rootcore = modeNproduct(rootcore,...
res.coeffs{i}/sum(res.coeffs{i})*...
Tensor.rootfactors{i+Tensor.numTechnicalModes-1},i+Tensor.numTechnicalModes-1);
end
res.motRec = createMotionFromCoreTensor_jt(core,rootcore,skel,true,true,Tensor.DataRep);
end
%%
function f = objfunWii(x,optimStruct)
X = mat2cell(x,1,optimStruct.tensor.dimNaturalModes)';
for i=1:optimStruct.tensor.numNaturalModes
optimStruct.preparedCore = modeNproduct(optimStruct.preparedCore,...
X{i}*optimStruct.tensor.factors{i+optimStruct.tensor.numTechnicalModes},...
i+optimStruct.tensor.numTechnicalModes);
optimStruct.preparedRootCore = modeNproduct(optimStruct.preparedRootCore,...
X{i}*optimStruct.tensor.rootfactors{i+optimStruct.tensor.numTechnicalModes-1},...
i+optimStruct.tensor.numTechnicalModes-1);
end
motRec = createMotionFromCoreTensor_jt(optimStruct.preparedCore,...
optimStruct.preparedRootCore,...
optimStruct.skel,...
true,...
true,...
optimStruct.tensor.DataRep);
motRec = addAccToMot(motRec);
recData = motRec.jointAccelerations{optimStruct.joint};
recData(2,:)=recData(2,:)+9.81;
motRec = computeLocalSystems(optimStruct.skel,motRec);
P = C_quatrot(recData,C_quatinv(motRec.localSystems{optimStruct.joint}));
Q = optimStruct.wiiData;
% [R,T] = icp(P,Q);
T = findOptimalPCtransformation(P,Q);
f = T.pc1_new - Q;
end
%%
function wii_cut = cutWiiData(wiidata,refdata)
wiiNorm = normOfColumns(wiidata);
refNorm = normOfColumns(refdata);
mindist = inf;
for i=1:length(wiiNorm)-length(refNorm)+1
T = findOptimalPCtransformation(wiidata(:,i:i+length(refdata)-1),refdata);
dist = sqrt(sum((T.pc1_new - refdata).^2));
% fprintf('%i\n',sum(dist));
if sum(dist)<mindist
mindist=sum(dist);
strt=i;
end
end
wii_cut = wiidata(:,strt:strt+length(refNorm)-1);
fprintf('\nstart frame: %i, end frame: %i\n',strt,strt+length(refNorm)-1);
plot(wiiNorm(:,strt:strt+length(refNorm)-1));
hold all;
plot(refNorm);
drawnow();
end
%%
function newdata = changeSamplingRate(data,oldRate,newRate)
oldSampling=1:length(data);
newSampling=1:oldRate/newRate:length(data);
newdata = spline(oldSampling,data,newSampling);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
translateMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/translateMotion.m
| 698 |
utf_8
|
33e6539750837edc3271f125a42f6ec7
|
% function translateMotion
% translates a motion with specified translation
% mot = translateMotion(skel,mot,x,y,z)
% author: Jochen Tautges ([email protected])
function mot = translateMotion(skel,mot,x,y,z)
mot.rootTranslation = mot.rootTranslation+repmat([x;y;z],1,mot.nframes);
mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
if (isfield(mot,'jointVelocities')&&~isempty(mot.jointVelocities))
mot=addVelToMot(mot);
end
if (isfield(mot,'jointAccelerations')&&~isempty(mot.jointAccelerations))
mot=addAccToMot(mot);
end
%fprintf('Motion successfully translated with x=%.2f, y=%.2f, z=%.2f.\n',x,y,z);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
mirrorMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/mirrorMot.m
| 1,554 |
utf_8
|
31552f5474f3dde982ed755a26923111
|
% function mirrorMot
% mirrors specified skeleton and motion on the yz-plane
% [newskel,newmot] = mirrorMot(skel,mot)
% author: Jochen Tautges ([email protected])
function [newskel,newmot]=mirrorMot(skel,mot)
newskel = mirrorSkel(skel);
% if isempty(mot.rotationEuler)
mot = C_convert2euler(skel,mot);
% end
newmot = mot;
newmot.rootTranslation(1,:) = -newmot.rootTranslation(1,:);
pairs{1} = [6,11];
pairs{2} = [5,10];
pairs{3} = [9,4];
pairs{4} = [3,8];
pairs{5} = [2,7];
pairs{6} = [18,25];
pairs{7} = [19,26];
pairs{8} = [20,27];
pairs{9} = [21,28];
pairs{10} = [22,29];
pairs{11} = [23,30];
pairs{12} = [24,31];
for i=1:length(pairs)
newmot.rotationEuler{pairs{i}(1)} = mot.rotationEuler{pairs{i}(2)};
newmot.rotationEuler{pairs{i}(2)} = mot.rotationEuler{pairs{i}(1)};
end
newmot.rotationEuler{1}(2,:) = -newmot.rotationEuler{1}(2,:);
newmot.rotationEuler{1}(3,:) = -newmot.rotationEuler{1}(3,:);
for i=2:mot.njoints
idx = strmatch('ry', lower(skel.nodes(i).DOF), 'exact');
if ~isempty(idx)
newmot.rotationEuler{i}(idx,:) = -newmot.rotationEuler{i}(idx,:);
end
idx = strmatch('rz', lower(skel.nodes(i).DOF), 'exact');
if ~isempty(idx)
newmot.rotationEuler{i}(idx,:) = -newmot.rotationEuler{i}(idx,:);
end
end
newmot = C_convert2quat(newskel,newmot);
newmot.jointTrajectories = iterativeForwKinematics(newskel,newmot);
newmot.boundingBox = computeBoundingBox(newmot);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
diff5point.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/diff5point.m
| 1,268 |
utf_8
|
d19f6076adaa6ec1abfbb4f8d9262385
|
% function newStream = diff5point(stream,samplingRate,varargin)
% 5-point derivation
% author: Jochen Tautges ([email protected])
function newStream = diff5point(stream,samplingRate,varargin)
switch nargin
case 2
weights = [1 -8 0 8 -1];
divisor = 12/samplingRate;
case 3
if varargin{1}==1
weights = [1 -8 0 8 -1];
divisor = 12/samplingRate;
elseif varargin{1}==2;
weights = [-1 16 -30 16 -1];
divisor = 12/samplingRate^2;
else
error('Not implemented yet!');
end
otherwise
error('Wrong number of argins');
end
newStream = zeros(size(stream));
% padding
stream = [ 3*stream(:,1)-2*stream(:,2),...
2*stream(:,1)-stream(:,2),...
stream,...
2*stream(:,end)-stream(:,end-1),...
3*stream(:,end)-2*stream(:,end-1)];
for i = 3:size(newStream,2)+2
newStream(:,i-2) = ...
weights(1) * stream(:,i-2) ...
+ weights(2) * stream(:,i-1) ...
+ weights(3) * stream(:,i) ...
+ weights(4) * stream(:,i+1) ...
+ weights(5) * stream(:,i+2);
end
newStream = newStream / divisor;
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
readNrOfFramesFromFile.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/readNrOfFramesFromFile.m
| 809 |
utf_8
|
516fa9a9fee29687bc51f14188778810
|
% function readNrOfFramesFroMFile
% reads number of frames from specified amc-file
% cf. function readNumFrames
function nrOfFrames = readNrOfFramesFromFile(filename)
fid = fopen(filename,'rt');
found = false; % found last frame number
fseek(fid,-1,'eof'); % jump to end of file
while (ftell(fid) > 0 && ~found) % not at bof AND not found last frame number
pos = ftell(fid);
ch = fread(fid,1,'uchar'); % read current character
if (ch == 10) % line feed
l = fgetl(fid);
fseek(fid,pos-1,'bof');
if (l ~= -1) % no eof encountered
nrOfFrames = str2double(l);
if ~isnan(nrOfFrames)
found = true;
end
end
else
fseek(fid,-2,'cof'); % jump 2 characters back
end
end
fclose(fid);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
getStepSizes.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/getStepSizes.m
| 4,160 |
utf_8
|
fb6790e712da00511e990048a8359fbc
|
% function getStepSizes
% (naively) computes stepSizes for walking motions
% stepSizes = getStepSizes(skel,mot)
% author: Jochen Tautges ([email protected])
function [stepSizesMin,stepSizesMax] = getStepSizes(mot)
leftJoint = 4;
rightJoint = 9;
% for i=1:mot.nframes-1
% if ((mot.jointTrajectories{leftJoint}(2,i)<=mot.jointTrajectories{rightJoint}(2,i))...
% && (mot.jointTrajectories{leftJoint}(2,i+1)>mot.jointTrajectories{rightJoint}(2,i+1)))...
% || ((mot.jointTrajectories{leftJoint}(2,i)>=mot.jointTrajectories{rightJoint}(2,i))...
% && (mot.jointTrajectories{leftJoint}(2,i+1)<mot.jointTrajectories{rightJoint}(2,i+1)))
%
% counter=counter+1;
% stepSizes(counter)=normOfColumns(mot.jointTrajectories{5}(:,i)-mot.jointTrajectories{10}(:,i));
%
% elseif ((i==mot.nframes-1)...
% && (mot.jointTrajectories{leftJoint}(2,i+1)==mot.jointTrajectories{rightJoint}(2,i+1)))
%
% counter=counter+1;
% stepSizes(counter)=normOfColumns(mot.jointTrajectories{5}(:,i+1)-mot.jointTrajectories{10}(:,i+1));
% end
% end
leftToRight=findstr([0 sign((mot.jointTrajectories{leftJoint}(2,:))-(mot.jointTrajectories{rightJoint}(2,:)))],[-1 1]);
rightToLeft=findstr([0 sign((mot.jointTrajectories{rightJoint}(2,:))-(mot.jointTrajectories{leftJoint}(2,:)))],[-1 1]);
frameIDs=union(1,[leftToRight,rightToLeft,mot.nframes]);
if mot.jointTrajectories{leftJoint}(2,1)<=mot.jointTrajectories{rightJoint}(2,1) % left foot on floor in first frame
joints=[leftJoint rightJoint];
else
joints=[rightJoint leftJoint];
end
maxPositions = zeros(1,length(frameIDs)-1);
minPositions = zeros(1,length(frameIDs)-1);
maxValues = zeros(3,length(frameIDs)-1);
minValues = zeros(3,length(frameIDs)-1);
for i=2:length(frameIDs)
[minValue,minPos]=min(mot.jointTrajectories{joints(1)}(2,frameIDs(i-1):frameIDs(i)));
[maxValue,maxPos]=max(mot.jointTrajectories{joints(2)}(2,frameIDs(i-1):frameIDs(i)));
maxPositions(i-1)=maxPos+frameIDs(i-1)-1;
minPositions(i-1)=minPos+frameIDs(i-1)-1;
minValues(:,i-1)=mot.jointTrajectories{joints(1)}(:,minPositions(i-1));
maxValues(:,i-1)=mot.jointTrajectories{joints(1)}(:,maxPositions(i-1)); % joints(1) is correct!
joints=fliplr(joints);
end
stepSizesMin = zeros(1,length(minValues)-1);
stepSizesMax = zeros(1,length(maxValues)-1);
for i=2:length(minValues)
stepSizesMin(i-1)=normOfColumns(minValues(:,i)-minValues(:,i-1));
end
for i=2:length(maxValues)
stepSizesMax(i-1)=normOfColumns(maxValues(:,i)-maxValues(:,i-1));
end
% lastOnFloor='';
% counter=0;
% minID=0;
% coordinates=[];
%
% for i=1:mot.nframes
% if mot.jointTrajectories{leftJoint}(2,i)<=mot.jointTrajectories{rightJoint}(2,i) % left foot on floor
% if ~strcmp(lastOnFloor,'left')
% counter=counter+1;
% minSoFar=inf;
% if minID~=0
% if ~isempty(coordinates)
% stepSizes(counter)=normOfColumns(coordinates-mot.jointTrajectories{rightJoint}(:,minID));
% end
% coordinates=mot.jointTrajectories{rightJoint}(:,minID);
% end
% end
% if mot.jointTrajectories{leftJoint}(2,i)<minSoFar
% minSoFar=mot.jointTrajectories{leftJoint}(2,i);
% minID=i;
% end
% lastOnFloor='left';
% else % right foot on Floor
% if ~strcmp(lastOnFloor,'right')
% counter=counter+1;
% minSoFar=inf;
% if minID~=0
% if ~isempty(coordinates)
% stepSizes(counter)=normOfColumns(coordinates-mot.jointTrajectories{leftJoint}(:,minID));
% end
% coordinates=mot.jointTrajectories{leftJoint}(:,minID);
% end
% end
% if mot.jointTrajectories{rightJoint}(2,i)<minSoFar
% minSoFar=mot.jointTrajectories{rightJoint}(2,i);
% minID=i;
% end
% lastOnFloor='right';
% end
% end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recMotWithPCAmat.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/recMotWithPCAmat.m
| 9,092 |
utf_8
|
8e2d52213cfa8169d8d9e2c0ae5f401f
|
% function recMotWithPCA(dataMat,skel,mot,dataRep)
%
% dataMat = dofs x frames
% skel = skel struct of original motion
% mot = mot struct of original motion
% dataRep = 'quat' or 'euler'
function res = recMotWithPCAmat(dataMat,skel,mot,dataRep)
mot.rootTranslation = zeros(3,mot.nframes);
mot = fitRootOrientationsFrameWise(skel,mot);
mot.boundingBox = computeBoundingBox(mot);
res.controlJoints = [4 9 20 27];
res.kdJoints = [3 4 8 9 19 20 26 27];
Xr = [];
switch dataRep
case 'euler'
optimStruct.dofs = [3 0 3 1 2 1 0 3 1 2 1 3 3 3 3 3 3 2 3 1 1 2 1 2 2 3 1 1 2 1 2];
case 'quat'
optimStruct.dofs = [4 0 4 4 4 4 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4];
end
dofsCum = [0 cumsum(optimStruct.dofs)];
for i=res.kdJoints
Xr = [Xr ; dataMat(dofsCum(i)+1:dofsCum(i+1),:)];
end
% --------------------------------------------
% maximum number of nearest neightbours: size(dataMat,2);
res.k = size(dataMat,2);
% maximum number of nearest neighbours for PCA: res.k;
% res.kForOptimization = 50;
% maximum number of principal components: size(dataMat,1);
res.nrOfPrinComps = 30;
res.nrOfPrinCompsForCovMat = 20;
% ---------------------------------------------
lb = [];
ub = [];
options = optimset( 'Display','iter',...
'MaxFunEvals',100000,...
'MaxIter',100,...
'TolFun',0.01);%,...
% 'PlotFcns', @optimplotx);
startFrame = 1;
% endFrame = mot.nframes;
endFrame = 20;
optimStruct.dataRep = dataRep;
optimStruct.q = cell(mot.nframes,1);
optimStruct.q{startFrame} = cutMotion(mot,startFrame,startFrame);
optimStruct.q{startFrame+1} = cutMotion(mot,startFrame+1,startFrame+1);
res.recMot = cutMotion(mot,startFrame,startFrame+1);
switch dataRep
case 'euler'
angles_curr = cell2mat(optimStruct.q{startFrame+1}.rotationEuler);
Xq = cell2mat(optimStruct.q{startFrame+1}.rotationEuler(res.kdJoints));
case 'quat'
angles_curr = cell2mat(optimStruct.q{startFrame+1}.rotationQuat);
Xq = cell2mat(optimStruct.q{startFrame+1}.rotationQuat(res.kdJoints));
end
[nnidx,dists] = annquery(Xr, Xq, res.k);
optimStruct.dataMatKD = dataMat(:,nnidx');
[optimStruct.coeffs,optimStruct.score] = princomp(optimStruct.dataMatKD');
optimStruct.meanVec = mean(optimStruct.dataMatKD,2);
% optimStruct.dataMatKD ==
% optimStruct.coeffs * optimStruct.score' + repmat(optimStruct.meanVec,1,res.k)
optimStruct.priorFactor = pinv(optimStruct.coeffs(:,1:res.nrOfPrinCompsForCovMat)')...
* inv(cov(optimStruct.score(:,1:res.nrOfPrinCompsForCovMat)))...
* pinv(optimStruct.coeffs(:,1:res.nrOfPrinCompsForCovMat));
% s = (dists(res.kForOptimization)*ones(res.kForOptimization,1)-dists(1:res.kForOptimization));
% w = s/sum(s);
% angles_curr = optimStruct.dataMatKD(:,1:res.kForOptimization)*w;
% startValue = (angles_curr-optimStruct.meanVec)' ...
% * pinv(optimStruct.coeffs(:,1:res.nrOfPrinComps)') ...
% * pinv(optimStruct.score(1:res.kForOptimization,1:res.nrOfPrinComps));
startValue = (angles_curr-optimStruct.meanVec)' ...
* pinv(optimStruct.coeffs(:,1:res.nrOfPrinComps)');
res.origMot = cutMotion(mot,startFrame,endFrame);
res.skel = skel;
for i=startFrame+2:endFrame
fprintf('\nReconstructing frame %i...\n',i);
optimStruct.frame = i;
optimStruct.q_orig = cutMotion(res.origMot,i,i);
X = lsqnonlin(@(x) objfunPCAlocal(x,res,optimStruct),startValue,lb,ub,options);
% angles_curr = (X * optimStruct.score(1:res.kForOptimization,1:res.nrOfPrinComps)...
% * optimStruct.coeffs(1:res.nrOfPrinComps,:)')'...
% + optimStruct.meanVec;
angles_curr = optimStruct.coeffs(:,1:res.nrOfPrinComps) * X'...
+ optimStruct.meanVec;
optimStruct.q{i} = buildMotFromAngles(angles_curr,skel,optimStruct.dofs,dataRep);
res.recMot = addFrame2Motion(res.recMot,optimStruct.q{i});
save('res','res');
switch dataRep
case 'euler'
Xq = cell2mat(optimStruct.q{i}.rotationEuler(res.kdJoints));
case 'quat'
Xq = cell2mat(optimStruct.q{i}.rotationQuat(res.kdJoints));
end
[nnidx,dists] = annquery(Xr, Xq, res.k);
optimStruct.dataMatKD = dataMat(:,nnidx');
[optimStruct.coeffs,optimStruct.score] = princomp(optimStruct.dataMatKD');
optimStruct.meanVec = mean(optimStruct.dataMatKD,2);
optimStruct.priorFactor = pinv(optimStruct.coeffs(:,1:res.nrOfPrinCompsForCovMat)')...
* inv(cov(optimStruct.score(:,1:res.nrOfPrinCompsForCovMat)))...
* pinv(optimStruct.coeffs(:,1:res.nrOfPrinCompsForCovMat));
% s = (dists(res.kForOptimization)*ones(res.kForOptimization,1)-dists(1:res.kForOptimization));
% w = s/sum(s);
% angles_curr = optimStruct.dataMatKD(:,1:res.kForOptimization)*w;
% startValue = ((angles_curr-optimStruct.meanVec)'...
% * pinv(optimStruct.coeffs(:,1:res.nrOfPrinComps)'))...
% * pinv(optimStruct.score(1:res.kForOptimization,1:res.nrOfPrinComps));
startValue = (angles_curr-optimStruct.meanVec)' ...
* pinv(optimStruct.coeffs(:,1:res.nrOfPrinComps)');
end
res.recMot.samplingRate = res.origMot.samplingRate;
end
%% function objfunPCAlocal
function f = objfunPCAlocal(x,res,optimStruct)
% angles_curr = (x * optimStruct.score(1:res.kForOptimization,1:res.nrOfPrinComps)...
% * optimStruct.coeffs(1:res.nrOfPrinComps,:)')'...
% + optimStruct.meanVec;
angles_curr = optimStruct.coeffs(:,1:res.nrOfPrinComps) * x'...
+ optimStruct.meanVec;
q_curr = buildMotFromAngles(angles_curr,res.skel,optimStruct.dofs,optimStruct.dataRep);
% ------- control term ----------
e_control = cell2mat(q_curr.jointTrajectories(res.controlJoints))...
- cell2mat(optimStruct.q_orig.jointTrajectories(res.controlJoints));
% e_control = mean(e_control.^2);
% ------- smoothness ------------
e_smooth = cell2mat(q_curr.jointTrajectories)...
- 2 * cell2mat(optimStruct.q{optimStruct.frame-1}.jointTrajectories)...
+ cell2mat(optimStruct.q{optimStruct.frame-2}.jointTrajectories);
% e_smooth = mean(e_smooth.^2);
% ------- prior -----------------
e_prior = (angles_curr-optimStruct.meanVec)'...
* optimStruct.priorFactor...
* (angles_curr-optimStruct.meanVec);
% e_prior = e_prior / length(angles_curr);
% ------- overall error --------
% weights
control_w = 1 / numel(e_control);
smooth_w = 1 / numel(e_smooth);
prior_w = 1 / numel(e_prior);
% f = control_w * e_control ...
% + smooth_w * e_smooth ...
% + prior_w * e_prior;
f = [control_w * e_control(:);...
smooth_w * e_smooth(:);...
prior_w * e_prior(:)];
% bar([control_w*sum(e_control.^2) smooth_w*sum(e_smooth.^2) prior_w*e_prior^2]);
% drawnow();
end
%% function buildMotFromAngles
function q_curr = buildMotFromAngles(angles_curr,skel,dofs,dataRep)
q_curr = emptyMotion;
q_curr.njoints = 31;
q_curr.nframes = 1;
q_curr.rootTranslation = [0;0;0];
switch dataRep
case 'euler'
q_curr.rotationEuler = mat2cell(angles_curr,dofs);
q_curr = convert2quat(skel,q_curr);
case 'quat'
q_curr.rotationQuat = mat2cell(angles_curr,dofs);
% q_curr = convert2euler(skel,q_curr);
end
q_curr.jointTrajectories = forwardKinematicsQuat(skel,q_curr);
q_curr.boundingBox = computeBoundingBox(q_curr);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
cutMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/cutMotion.m
| 3,166 |
utf_8
|
22b84462b6dea40f66c0315e91e41e86
|
% function cutMotion
% cuts a motion sequence to specified start and end frame
% mot = cutMotion(mot,startFrame,endFrame)
% authors: Bjoern Krueger ([email protected]),
% Jochen Tautges ([email protected])
function [mot] = cutMotion(mot,startF,endF)
if startF~=1 || endF~=mot.nframes
if (startF<1 || startF>endF || startF>mot.nframes || endF>mot.nframes)
error('Forbidden values for start and end frame! sF=%i eF=%i mot.nframes=%i',startF, endF, mot.nframes);
end
mot.nframes = endF-startF+1;
% computePos = false;
if (isfield(mot,'jointTrajectories') && ~isempty(mot.jointTrajectories))
% computePos = true;
if iscell(mot.jointTrajectories)
if size(mot.jointTrajectories{1},2)>=endF
mot.jointTrajectories = cellfun(@(x) x(:,startF:endF),mot.jointTrajectories,'UniformOutput',0);
end
mot.boundingBox = computeBoundingBox(mot);
else
mot.jointTrajectories = mot.jointTrajectories(:,startF:endF);
end
end
% computeVel = false;
if (isfield(mot,'jointVelocities') && ~isempty(mot.jointVelocities))
if iscell(mot.jointVelocities)
mot.jointVelocities = cellfun(@(x) x(:,startF:endF),mot.jointVelocities,'UniformOutput',0);
else
mot.jointVelocities = mot.jointVelocities(:,startF:endF);
end
% computeVel = true;
end
% computeAcc = false;
if (isfield(mot,'jointAccelerations')&& ~isempty(mot.jointAccelerations))
% computeAcc = true;
if iscell(mot.jointAccelerations)
mot.jointAccelerations = cellfun(@(x) x(:,startF:endF),mot.jointAccelerations,'UniformOutput',0);
else
mot.jointAccelerations = mot.jointAccelerations(:,startF:endF);
end
end
% computeQuat = false;
if (isfield(mot,'rotationQuat') && ~isempty(mot.rotationQuat))
if(iscell(mot.rotationQuat))
% computeQuat = true;
mot.rotationQuat(mot.animated) = cellfun(@(x) x(:,startF:endF),mot.rotationQuat(mot.animated),'UniformOutput',0);
mot.rotationQuat(mot.unanimated) = {[]};
else
mot.rotationQuat = mot.rotationQuat(:,startF:endF);
end
end
% computeEuler = false;
if (isfield(mot,'rotationEuler') && ~isempty(mot.rotationEuler))
% computeEuler = true;
mot.rotationEuler(mot.animated) = cellfun(@(x) x(:,startF:endF),mot.rotationEuler(mot.animated),'UniformOutput',0);
mot.rotationEuler(mot.unanimated) = {[]};
end
if ~isempty(mot.rootTranslation)
mot.rootTranslation = mot.rootTranslation(:,startF:endF);
end
if isfield(mot,'footprints')
mot.footprints = mot.footprints(:,startF:endF);
end
if isfield(mot,'timestamp')
mot.timestamp = mot.timestamp(:,startF:endF);
end
% crazy people special field
if (isfield(mot,'rootOri') && ~isempty(mot.rootOri))
mot.rootOri = mot.rootOri(:,startF:endF);
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
mirrorSkel.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/mirrorSkel.m
| 1,411 |
utf_8
|
083463bbc71dba379e79c0608dae419f
|
% function mirrorSkel
% mirrors specified skeleton on the yz-plane
% newskel = mirrorSkel(skel)
% author: Jochen Tautges ([email protected])
function newskel=mirrorSkel(skel)
newskel = skel;
newskel.filename = [skel.filename '.mirrored'];
pairs{1} = [6,11];
pairs{2} = [5,10];
pairs{3} = [9,4];
pairs{4} = [3,8];
pairs{5} = [2,7];
pairs{6} = [18,25];
pairs{7} = [19,26];
pairs{8} = [20,27];
pairs{9} = [21,28];
pairs{10} = [22,29];
pairs{11} = [23,30];
pairs{12} = [24,31];
for i=1:length(pairs)
newskel.nodes(pairs{i}(1)).length = skel.nodes(pairs{i}(2)).length;
newskel.nodes(pairs{i}(2)).length = skel.nodes(pairs{i}(1)).length;
newskel.nodes(pairs{i}(1)).offset = skel.nodes(pairs{i}(2)).offset;
newskel.nodes(pairs{i}(2)).offset = skel.nodes(pairs{i}(1)).offset;
newskel.nodes(pairs{i}(1)).direction = skel.nodes(pairs{i}(2)).direction;
newskel.nodes(pairs{i}(2)).direction = skel.nodes(pairs{i}(1)).direction;
newskel.nodes(pairs{i}(1)).axis = -skel.nodes(pairs{i}(2)).axis;
newskel.nodes(pairs{i}(2)).axis = -skel.nodes(pairs{i}(1)).axis;
end
for i=1:skel.njoints
newskel.nodes(i).offset(1) = -newskel.nodes(i).offset(1);
newskel.nodes(i).direction(1) = -newskel.nodes(i).direction(1);
newskel.nodes(i).axis(1) = -newskel.nodes(i).axis(1);
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recMotFromAcc.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/recMotFromAcc.m
| 4,581 |
utf_8
|
9f50d3e6824a1923102a16358de9bac5
|
% function recMotFromAcc
%
% db.acc
% db.quat
%
% open questions concerning db:
% - coordinate frame transformation
% - resampling
% - sensor/joint offsets
% - flips that screw up accelerations
%
% more unsolved problems
% - initialization (t-pose)
% - choice of skeleton (optimization?)
function res = recMotFromAcc(db,skel,mot)
%% optional settings --------------------------------------------------------
nrOfNN = 1;
eps = 0.1;
% radius = 5;
res.joints = [5,10,21,28];
acc_direction = 'xyz';
newFrameRate = db.frameRate;
startFrame = 1;
endFrame = mot.nframes;
res.windowSize = 16;
%% preprocessing ----------------------------------------------------------
fprintf('Preprocessing:\n');
res.joints = sort(res.joints);
idx = jointIDsToMatrixIndices(res.joints);
nrOfJoints = numel(res.joints);
acc_idx = [];
if sum(acc_direction=='x')~=0, acc_idx=[acc_idx 1:3:nrOfJoints*3]; end
if sum(acc_direction=='y')~=0, acc_idx=[acc_idx 2:3:nrOfJoints*3]; end
if sum(acc_direction=='z')~=0, acc_idx=[acc_idx 3:3:nrOfJoints*3]; end
acc_idx = sort(acc_idx);
% % modification of original motion
% fprintf('Modification of original motion started...'); tic;
% mot = cutMotion(mot,startFrame,endFrame);
% mot = changeFrameRate(skel,mot,newFrameRate);
% res.origmot = mot;
% mot.rootTranslation(:,:) = 0;
% mot = fitRootOrientationsFrameWise(skel,mot);
% if ~isfield(mot,'jointAcceleration');
% mot = addAccToMot(mot);
% end
% mot.boundingBox = computeBoundingBox(mot);
% res.origmot_mod = mot;
% fprintf(' finished in %.2f seconds.\n',toc);
mot = cutMotion(mot,startFrame,endFrame);
mot = changeFrameRate(skel,mot,newFrameRate);
res.origmot_mod = mot;
% tree construction
treeData = zeros(nrOfJoints*res.windowSize*length(acc_direction),size(db.pos,2)-res.windowSize+1);
fprintf('kd-tree construction started (#frames = %i, #dims = %i = %i joints x %i accs x %i frames)...\n',...
size(treeData'), nrOfJoints,length(acc_direction),res.windowSize); tic;
% treeHandle = ann_buildTree(double(db.acc(idx.pos,:)));
for i=1:size(db.acc,2)-res.windowSize+1
tmp = db.acc(idx.pos(acc_idx),i:i+res.windowSize-1)';
treeData(:,i) = tmp(:);
end
treeHandle = ann_buildTree(treeData);
fprintf('finished in %.2f seconds.\n',toc);
% treeQuery = cell2mat(mot.jointAccelerations(res.joints));
ma = cell2mat(mot.jointAccelerations(res.joints));
ma = ma(acc_idx,:);
treeQuery = zeros(nrOfJoints*res.windowSize*length(acc_direction),floor(mot.nframes/res.windowSize));
for i=1:res.windowSize:mot.nframes-res.windowSize+1
tmp = ma(:,i:i+res.windowSize-1)';
treeQuery(:,(i-1)/res.windowSize+1) = tmp(:);
end
fprintf('nn-search started (#frames = %i, #neighbours = %i)...',mot.nframes,nrOfNN);tic;
% [nnidx,nndists] = ann_queryTree(treeHandle,treeQuery,nrOfNN,'eps',eps,'search_sch','fr','radius',radius);
[res.nnidx,res.nndists] = ann_queryTree(treeHandle,treeQuery,nrOfNN,'eps',eps);
fprintf(' finished in %.2f seconds.\n',toc);
%% reconstruction ---------------------------------------------------------
fprintf('Reconstruction:\n');
hit = 1;
fprintf('Concatenating hits of rank %i and length %i...', hit, res.windowSize);
res.recmot = buildMotFromRotData(db.quat(:,res.nnidx(hit,1):res.nnidx(hit,1)+res.windowSize-1),skel);
res.recmot.samplingRate = 30;
res.recmot.frameTime = 1/30;
res.recmot.jointAccelerations = mat2cell(db.acc(:,res.nnidx(hit,1):res.nnidx(hit,1)+res.windowSize-1),3*ones(1,31),res.windowSize);
% res.recmot = addAccToMot(res.recmot);
for i=2:size(res.nnidx,2)
tmp_mot = buildMotFromRotData(db.quat(:,res.nnidx(hit,i):res.nnidx(hit,i)+res.windowSize-1),skel);
tmp_mot.samplingRate = 30;
tmp_mot.frameTime = 1/30;
tmp_mot.jointAccelerations = mat2cell(db.acc(:,res.nnidx(hit,i):res.nnidx(hit,i)+res.windowSize-1),3*ones(1,31),res.windowSize);
% tmp_mot = addAccToMot(tmp_mot);
res.recmot = addFrame2Motion(res.recmot,tmp_mot);
end
fprintf(' finished in %.2f seconds.\n',toc);
%% postprocessing ---------------------------------------------------------
fprintf('Postprocessing:\n');
tic;
ann_cleanup;
fprintf('\b in %.2f seconds.\n',toc);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructCMUmotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/reconstructCMUmotion.m
| 1,714 |
utf_8
|
5496ac3f901d2b3599492c4a250f9c60
|
% res=reconstructCMUmotion(TensorForOptimization,CMUresult[,joints[,TensorForReconstruction]]])
function res = reconstructCMUmotion(TensorForReconstruction,skel,origmot,varargin)
switch nargin
case 3
joints = (1:31);
case 4
joints = varargin{1};
case 5
joints = varargin{1};
TensorForReconstruction=varargin{2};
otherwise
('Wrong number of argins!');
end
% folder='\\breithorn\shareII\MoCap-Daten\CMU_DB\CMU_D180\AMC\locomotion\walking\';
%
% asfName = strcat(folder,CMUresult.asf);
% res.skel = readASF(asfName);
% amcName = strcat(folder,CMUresult.amc(max(strfind(CMUresult.amc,'/'))+1:end));
% origmot = readAMC(amcName,res.skel);
% res.skel = CMUresult.orgSkel;
% origmot = CMUresult.orgMot;
res.skel = skel;
res.origmot = changeFrameRate(skel,origmot,30);
[mot,angle,x0,z0] = fitMotion(res.skel,res.origmot);
[s,motRef] = extractMotion(TensorForReconstruction,[1,1,1]);
[D,warpingPath] = pointCloudDTW_jt(motRef,mot);
mot = warpMotion(warpingPath,res.skel,mot);
set = defaultSet;
set.regardedJoints = joints;
[res.X,recmot] = reconstructMotion(TensorForReconstruction,res.skel,mot,'set',set);
if nargin==5
[s,recmot] = constructMotion(TensorForReconstruction,res.X,res.skel);
end
recmot = warpMotion(fliplr(warpingPath),res.skel,recmot);
recmot = rotateMotion(res.skel,recmot,-angle,'y');
res.recmot = translateMotion(res.skel,recmot,-x0,0,-z0);
% recmot = removeSkating(skel,recmot);
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
zupt.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/zupt.m
| 3,001 |
utf_8
|
23a157324073343dc07482c998979a90
|
% zero velocity updates
% input:
% a - global accelerations (nSamples x 3), gravity removed
% t0 - start frame of motion (here velocity is said to be zero)
% T - end frame of motion (here velocity is said to be zero)
% frameRate - frame rate
% output:
% res.v - velocity before correction
% res.vc - velocity after correction
% res.x - position before correction
% res.xc - position after correction
%
% author: Jochen Tautges ([email protected])
function res = zupt(a,t0,T,frameRate)
% a(:,1)=a(:,1)-mean(a([1:t0 T:end],1));
% a(:,2)=a(:,2)-mean(a([1:t0 T:end],2));
% a(:,3)=a(:,3)-mean(a([1:t0 T:end],3));
offsetX_left = mean(a(1:t0,1));
offsetY_left = mean(a(1:t0,2));
offsetZ_left = mean(a(1:t0,3));
offsetX_right = mean(a(T:end,1));
offsetY_right = mean(a(T:end,2));
offsetZ_right = mean(a(T:end,3));
offsetX = offsetX_left:(-offsetX_left+offsetX_right)/(T-t0):offsetX_right;
offsetY = offsetY_left:(-offsetY_left+offsetY_right)/(T-t0):offsetY_right;
offsetZ = offsetZ_left:(-offsetZ_left+offsetZ_right)/(T-t0):offsetZ_right;
res.a = a(t0:T,:);
res.a(:,1)=res.a(:,1)-offsetX';
res.a(:,2)=res.a(:,2)-offsetY';
res.a(:,3)=res.a(:,3)-offsetZ';
T = T-t0+1;
t0 = 1;
nSamples = size(res.a,1);
res.v = cumtrapz(res.a);
res.vc = zeros(size(res.a));
for t = 1:nSamples
res.vc(t,:) = res.v(t,:) + (res.v(t0,:)-res.v(T,:))*(t/T) - res.v(t0,:);
end
res.x = cumtrapz(res.v);
res.xc = zeros(size(res.a));
for t = 1:nSamples
res.xc(t,:) = res.x(t,:) + (res.v(t0,:)-res.v(T,:))*(t^2)/(2*T) - res.v(t0,:)*t - res.x(t0,:);
end
res.v = res.v / frameRate;
res.vc = res.vc / frameRate;
res.x = res.x / frameRate^2;
res.xc = res.xc / frameRate^2;
% nSamples = size(a,1);
%
% res.a = a;
%
% res.v = cumtrapz(a);
% res.vc = zeros(size(a));
%
% for t = 1:nSamples
% res.vc(t,:) = res.v(t,:) + (res.v(t0,:)-res.v(T,:))*((t-t0+1)/(T-t0+1)) - res.v(t0,:);
% end
%
% res.x = cumtrapz(res.v);
% res.xc = zeros(size(a));
%
% for t = 1:nSamples
% res.xc(t,:) = res.x(t,:) + (res.v(t0,:)-res.v(T,:))*((t-t0+1)^2)/(2*(T-t0+1)) - res.v(t0,:)*(t-t0+1) - res.x(t0,:);
% end
%
% res.v = res.v / frameRate;
% res.vc = res.vc / frameRate;
% res.x = res.x / frameRate^2;
% res.xc = res.xc / frameRate^2;
figure;
subplot(2,4,1);
plot(res.a);
title('Global Accelerations');
subplot(2,4,2);
plot(res.v);
title('Velocities without correction');
subplot(2,4,3);
plot(res.x);
title('Positions without correction');
subplot(2,4,4);
plot3(res.x(:,1),res.x(:,2),res.x(:,3));
axis equal;
title('3D-positions without correction');
subplot(2,4,5);
plot(res.a);
title('Global accelerations');
subplot(2,4,6);
plot(res.vc);
title('Velocities with correction');
subplot(2,4,7);
plot(res.xc);
title('Positions with correction');
subplot(2,4,8);
plot3(res.xc(:,1),res.xc(:,2),res.xc(:,3));
axis equal;
title('3d-positions with correction');
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
editMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/editMotion.m
| 2,216 |
utf_8
|
724f383800cf40f393dfd8ffa43699d6
|
% constraints.jointID
% constraints.jointList
% constraints.pos
% constraints.frame
% constraints.windowsize
function mot = editMotion(skel,mot,constraints)
A = []; b = []; Aeq = []; beq = [];
lb = [];
ub = [];
for i=constraints.jointList'
lb = [lb; skel.nodes(i).limits(:,1)];
ub = [ub; skel.nodes(i).limits(:,2)];
end
options=optimset('Display','iter',...
'MaxIter', 10,...
'TolFun',0.1,...
'TolCon',0.1,...
'Algorithm','active-set');
dofs = getDOFsFromSkel(skel);
regardedDOFs = dofs.euler(constraints.jointList);
x0 = zeros(sum(regardedDOFs),1);
frame = cutMotion(mot,constraints.frame,constraints.frame);
X = fmincon(@objfun_local,x0,A,b,Aeq,beq,lb,ub,@(x) nonlcon_local(x,constraints,regardedDOFs,skel,frame),options);
y = [zeros(size(X,1),1) zeros(size(X,1),1) X zeros(size(X,1),1) zeros(size(X,1),1)];
x = [1 2 constraints.windowsize+1 2*constraints.windowsize 2*constraints.windowsize+1];
xx = 1:2*constraints.windowsize+1;
yy = spline(x,y,xx);
eulers = cell2mat(cellfun(@(x) x(:,constraints.frame-constraints.windowsize:constraints.frame+constraints.windowsize),...
mot.rotationEuler(constraints.jointList),'UniformOutput',0));
eulers = eulers + yy;
eulers = mat2cell(eulers,regardedDOFs,size(eulers,2));
c=0;
for i=constraints.jointList'
c=c+1;
mot.rotationEuler{i}(:,constraints.frame-constraints.windowsize:constraints.frame+constraints.windowsize)=eulers{c};
end
mot = convert2quat(skel,mot);
mot.jointTrajectories = forwardKinematicsQuat(skel,mot);
end
%% local functions
function F = objfun_local(x)
S = eye(size(x,1));
F = 1/2 * x' * S * x;
end
function [c,ceq] = nonlcon_local(x,constraints,regardedDOFs,skel,frame)
newEulers = cell2mat(frame.rotationEuler(constraints.jointList)) + x;
frame.rotationEuler(constraints.jointList) = mat2cell(newEulers,regardedDOFs,1);
frame = convert2quat(skel,frame);
frame.jointTrajectories = forwardKinematicsQuat(skel,frame);
ceq = frame.jointTrajectories{constraints.jointID}-constraints.pos;
c=[];
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
uipickfiles.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/uipickfiles.m
| 23,893 |
utf_8
|
fa44e011ea5d774cceebf0f15764edd5
|
function out = uipickfiles(varargin)
%uipickfiles: GUI program to select file(s) and/or directories.
%
% Syntax:
% files = uipickfiles('PropertyName',PropertyValue,...)
%
% The current directory can be changed by operating in the file navigator:
% double-clicking on a directory in the list to move further down the tree,
% using the popup menu to move up the tree, typing a path in the box to
% move to any directory or right-clicking on the path box to revisit a
% previously-listed directory.
%
% Files can be added to the list by double-clicking or selecting files
% (non-contiguous selections are possible with the control key) and
% pressing the Add button. Files in the list can be removed or re-ordered.
% When finished, a press of the Done button will return the full paths to
% the selected files in a cell array, structure array or character array.
% If the Cancel button is pressed then zero is returned.
%
% The following optional property/value pairs can be specified as arguments
% to control the indicated behavior:
%
% Property Value
% ---------- ----------------------------------------------------------
% FilterSpec String to specify starting directory and/or file filter.
% Ex: 'C:\bin' will start up in that directory. '*.txt'
% will list only files ending in '.txt'. 'c:\bin\*.txt' will
% do both. Default is to start up in the current directory
% and list all files. Can be changed with the GUI.
%
% REFilter String containing a regular expression used to filter the
% file list. Ex: '\.m$|\.mat$' will list files ending in
% '.m' and '.mat'. Default is empty string. Can be used
% with FilterSpec and both filters are applied. Can be
% changed with the GUI.
%
% Prompt String containing a prompt appearing in the title bar of
% the figure. Default is 'Select files'.
%
% NumFiles Scalar or vector specifying number of files that must be
% selected. A scalar specifies an exact value; a two-element
% vector can be used to specify a range, [min max]. The
% function will not return unless the specified number of
% files have been chosen. Default is [] which accepts any
% number of files.
%
% Output String specifying the data type of the output: 'cell',
% 'struct' or 'char'. Specifying 'cell' produces a cell
% array of strings, the strings containing the full paths of
% the chosen files. 'Struct' returns a structure array like
% the result of the dir function except that the 'name' field
% contains a full path instead of just the file name. 'Char'
% returns a character array of the full paths. This is most
% useful when you have just one file and want it in a string
% instead of a cell array containing just one string. The
% default is 'cell'.
%
% All properties and values are case-insensitive and need only be
% unambiguous. For example,
%
% files = uipickfiles('num',1,'out','ch')
%
% is valid usage.
% Version: 1.0, 25 April 2006
% Author: Douglas M. Schwarz
% Email: dmschwarz=ieee*org, dmschwarz=urgrad*rochester*edu
% Real_email = regexprep(Email,{'=','*'},{'@','.'})
% Define properties and set default values.
prop.filterspec = '*';
prop.refilter = '';
prop.prompt = 'Select files';
prop.numfiles = [];
prop.output = 'cell';
% Process inputs and set prop fields.
properties = fieldnames(prop);
arg_index = 1;
while arg_index <= nargin
arg = varargin{arg_index};
if ischar(arg)
prop_index = find(strncmpi(arg,properties,length(arg)));
if length(prop_index) == 1
prop.(properties{prop_index}) = varargin{arg_index + 1};
else
error('Property ''%s'' does not exist or is ambiguous.',arg)
end
arg_index = arg_index + 2;
elseif isstruct(arg)
arg_fn = fieldnames(arg);
for i = 1:length(arg_fn)
prop_index = find(strncmpi(arg_fn{i},properties,...
length(arg_fn{i})));
if length(prop_index) == 1
prop.(properties{prop_index}) = arg.(arg_fn{i});
else
error('Property ''%s'' does not exist or is ambiguous.',...
arg_fn{i})
end
end
arg_index = arg_index + 1;
else
error(['Properties must be specified by property/value pairs',...
' or structures.'])
end
end
% Validate FilterSpec property.
if isempty(prop.filterspec)
prop.filterspec = '*';
end
if ~ischar(prop.filterspec)
error('FilterSpec property must contain a string.')
end
% Validate REFilter property.
if ~ischar(prop.refilter)
error('REFilter property must contain a string.')
end
% Validate Prompt property.
if ~ischar(prop.prompt)
error('Prompt property must contain a string.')
end
% Validate NumFiles property.
if numel(prop.numfiles) > 2 || any(prop.numfiles < 0)
error('NumFiles must be empty, a scalar or two-element vector.')
end
prop.numfiles = unique(prop.numfiles);
if isequal(prop.numfiles,1)
numstr = 'Select exactly 1 file.';
elseif length(prop.numfiles) == 1
numstr = sprintf('Select exactly %d files.',prop.numfiles);
else
numstr = sprintf('Select %d to %d files.',prop.numfiles);
end
% Validate Output property.
legal_outputs = {'cell','struct','char'};
out_idx = find(strncmpi(prop.output,legal_outputs,length(prop.output)));
if length(out_idx) == 1
prop.output = legal_outputs{out_idx};
else
error(['Value of ''Output'' property, ''%s'', is illegal or '...
'ambiguous.'],prop.output)
end
% Initialize file lists.
[current_dir,f,e] = fileparts(prop.filterspec);
filter = [f,e];
if isempty(current_dir)
current_dir = pwd;
end
if isempty(filter)
filter = '*';
end
re_filter = prop.refilter;
full_filter = fullfile(current_dir,filter);
path_cell = path2cell(current_dir);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
% Initialize some data.
file_picks = {};
full_file_picks = {};
dir_picks = dir(' '); % Create empty directory structure.
show_full_path = false;
nodupes = true;
history = {current_dir};
% Create figure.
gray = get(0,'DefaultUIControlBackgroundColor');
fig = figure('Position',[0 0 740 445],...
'Color',gray,...
'WindowStyle','modal',...
'Resize','off',...
'NumberTitle','off',...
'Name',prop.prompt,...
'IntegerHandle','off',...
'CloseRequestFcn',@cancel,...
'CreateFcn',{@movegui,'center'});
% Create uicontrols.
uicontrol('Style','frame',...
'Position',[255 260 110 70])
uicontrol('Style','frame',...
'Position',[275 135 110 100])
navlist = uicontrol('Style','listbox',...
'Position',[10 10 250 320],...
'String',filenames,...
'Value',[],...
'BackgroundColor','w',...
'Callback',@clicknav,...
'Max',2);
pickslist = uicontrol('Style','listbox',...
'Position',[380 10 350 320],...
'String',{},...
'BackgroundColor','w',...
'Callback',@clickpicks,...
'Max',2);
openbut = uicontrol('Style','pushbutton',...
'Position',[270 300 80 20],...
'String','Open',...
'Enable','off',...
'Callback',@open);
arrow = [2 2 2 2 2 2 2 2 1 2 2 2;...
2 2 2 2 2 2 2 2 2 0 2 2;...
2 2 2 2 2 2 2 2 2 2 0 2;...
0 0 0 0 0 0 0 0 0 0 0 0;...
2 2 2 2 2 2 2 2 2 2 0 2;...
2 2 2 2 2 2 2 2 2 0 2 2;...
2 2 2 2 2 2 2 2 1 2 2 2];
arrow(arrow == 2) = NaN;
arrow_im = NaN*ones(16,76);
arrow_im(6:12,45:56) = arrow/2;
im = repmat(arrow_im,[1 1 3]);
addbut = uicontrol('Style','pushbutton',...
'Position',[270 270 80 20],...
'String','Add ',...
'Enable','off',...
'CData',im,...
'Callback',@add);
removebut = uicontrol('Style','pushbutton',...
'Position',[290 205 80 20],...
'String','Remove',...
'Enable','off',...
'Callback',@remove);
moveupbut = uicontrol('Style','pushbutton',...
'Position',[290 175 80 20],...
'String','Move Up',...
'Enable','off',...
'Callback',@moveup);
movedownbut = uicontrol('Style','pushbutton',...
'Position',[290 145 80 20],...
'String','Move Down',...
'Enable','off',...
'Callback',@movedown);
uicontrol('Position',[10 380 250 16],...
'Style','text',...
'String','Current Directory',...
'HorizontalAlignment','center')
dir_popup = uicontrol('Style','popupmenu',...
'Position',[10 335 250 20],...
'BackgroundColor','w',...
'String',path_cell(end:-1:1),...
'Value',1,...
'Callback',@dirpopup);
hist_cm = uicontextmenu;
pathbox = uicontrol('Position',[10 360 250 20],...
'Style','edit',...
'BackgroundColor','w',...
'String',current_dir,...
'HorizontalAlignment','left',...
'Callback',@change_path,...
'UIContextMenu',hist_cm);
hist_menus = [];
hist_cb = @history_cb;
hist_menus = make_history_cm(hist_cb,hist_cm,hist_menus,history);
uicontrol('Position',[10 425 80 16],...
'Style','text',...
'String','File Filter',...
'HorizontalAlignment','left')
uicontrol('Position',[100 425 160 16],...
'Style','text',...
'String','Reg. Exp. Filter',...
'HorizontalAlignment','left')
showallfiles = uicontrol('Position',[270 405 100 20],...
'Style','checkbox',...
'String','Show All Files',...
'Value',0,...
'HorizontalAlignment','left',...
'Callback',@togglefilter);
filter_ed = uicontrol('Position',[10 405 80 20],...
'Style','edit',...
'BackgroundColor','w',...
'String',filter,...
'HorizontalAlignment','left',...
'Callback',@setfilspec);
refilter_ed = uicontrol('Position',[100 405 160 20],...
'Style','edit',...
'BackgroundColor','w',...
'String',re_filter,...
'HorizontalAlignment','left',...
'Callback',@setrefilter);
viewfullpath = uicontrol('Style','checkbox',...
'Position',[380 335 230 20],...
'String','Show full paths',...
'Value',show_full_path,...
'HorizontalAlignment','left',...
'Callback',@showfullpath);
remove_dupes = uicontrol('Style','checkbox',...
'Position',[380 360 230 20],...
'String','Remove duplicates (as per full path)',...
'Value',nodupes,...
'HorizontalAlignment','left',...
'Callback',@removedupes);
uicontrol('Position',[380 405 350 20],...
'Style','text',...
'String','Selected Files',...
'HorizontalAlignment','center')
uicontrol('Position',[280 80 80 30],'String','Done',...
'Callback',@done);
uicontrol('Position',[280 30 80 30],'String','Cancel',...
'Callback',@cancel);
if ~isempty(prop.numfiles)
uicontrol('Position',[380 385 350 16],...
'Style','text',...
'String',numstr,...
'ForegroundColor','r',...
'HorizontalAlignment','center')
end
set(fig,'HandleVisibility','off')
uiwait(fig)
% Compute desired output.
switch prop.output
case 'cell'
out = full_file_picks;
case 'struct'
out = dir_picks(:);
case 'char'
out = char(full_file_picks);
case 'cancel'
out = 0;
end
% -------------------- Callback functions --------------------
function add(varargin)
values = get(navlist,'Value');
for i = 1:length(values)
dir_pick = fdir(values(i));
pick = dir_pick.name;
pick_full = fullfile(current_dir,pick);
dir_pick.name = pick_full;
if ~nodupes || ~any(strcmp(full_file_picks,pick_full))
file_picks{end + 1} = pick;
full_file_picks{end + 1} = pick_full;
dir_picks(end + 1) = dir_pick;
end
end
if show_full_path
set(pickslist,'String',full_file_picks,'Value',[]);
else
set(pickslist,'String',file_picks,'Value',[]);
end
set([removebut,moveupbut,movedownbut],'Enable','off');
end
function remove(varargin)
values = get(pickslist,'Value');
file_picks(values) = [];
full_file_picks(values) = [];
dir_picks(values) = [];
top = get(pickslist,'ListboxTop');
num_above_top = sum(values < top);
top = top - num_above_top;
num_picks = length(file_picks);
new_value = min(min(values) - num_above_top,num_picks);
if num_picks == 0
new_value = [];
set([removebut,moveupbut,movedownbut],'Enable','off')
end
if show_full_path
set(pickslist,'String',full_file_picks,'Value',new_value,...
'ListboxTop',top)
else
set(pickslist,'String',file_picks,'Value',new_value,...
'ListboxTop',top)
end
end
function open(varargin)
values = get(navlist,'Value');
if fdir(values).isdir
if strcmp(fdir(values).name,'.')
return
elseif strcmp(fdir(values).name,'..')
set(dir_popup,'Value',min(2,length(path_cell)))
dirpopup();
return
end
current_dir = fullfile(current_dir,fdir(values).name);
history{end+1} = current_dir;
history = unique(history);
hist_menus = make_history_cm(hist_cb,hist_cm,hist_menus,...
history);
full_filter = fullfile(current_dir,filter);
path_cell = path2cell(current_dir);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(dir_popup,'String',path_cell(end:-1:1),'Value',1)
set(pathbox,'String',current_dir)
set(navlist,'ListboxTop',1,'Value',[],'String',filenames)
set(addbut,'Enable','off')
set(openbut,'Enable','off')
end
end
function clicknav(varargin)
value = get(navlist,'Value');
nval = length(value);
dbl_click_fcn = @add;
switch nval
case 0
set([addbut,openbut],'Enable','off')
case 1
set(addbut,'Enable','on');
if fdir(value).isdir
set(openbut,'Enable','on')
dbl_click_fcn = @open;
else
set(openbut,'Enable','off')
end
otherwise
set(addbut,'Enable','on')
set(openbut,'Enable','off')
end
if strcmp(get(fig,'SelectionType'),'open')
dbl_click_fcn();
end
end
function clickpicks(varargin)
value = get(pickslist,'Value');
if isempty(value)
set([removebut,moveupbut,movedownbut],'Enable','off')
else
set(removebut,'Enable','on')
if min(value) == 1
set(moveupbut,'Enable','off')
else
set(moveupbut,'Enable','on')
end
if max(value) == length(file_picks)
set(movedownbut,'Enable','off')
else
set(movedownbut,'Enable','on')
end
end
if strcmp(get(fig,'SelectionType'),'open')
remove();
end
end
function dirpopup(varargin)
value = get(dir_popup,'Value');
len = length(path_cell);
path_cell = path_cell(1:end-value+1);
if ispc && value == len
current_dir = '';
full_filter = filter;
fdir = struct('name',getdrives,'date',datestr(now),...
'bytes',0,'isdir',1);
else
current_dir = cell2path(path_cell);
history{end+1} = current_dir;
history = unique(history);
hist_menus = make_history_cm(hist_cb,hist_cm,hist_menus,...
history);
full_filter = fullfile(current_dir,filter);
fdir = filtered_dir(full_filter,re_filter);
end
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(dir_popup,'String',path_cell(end:-1:1),'Value',1)
set(pathbox,'String',current_dir)
set(navlist,'String',filenames,'Value',[])
set(addbut,'Enable','off')
end
function change_path(varargin)
proposed_path = get(pathbox,'String');
% Process any directories named '..'.
proposed_path_cell = path2cell(proposed_path);
ddots = strcmp(proposed_path_cell,'..');
ddots(find(ddots) - 1) = true;
proposed_path_cell(ddots) = [];
proposed_path = cell2path(proposed_path_cell);
% Check for existance of directory.
if ~exist(proposed_path,'dir')
uiwait(errordlg(['Directory "',proposed_path,...
'" does not exist.'],'','modal'))
return
end
current_dir = proposed_path;
history{end+1} = current_dir;
history = unique(history);
hist_menus = make_history_cm(hist_cb,hist_cm,hist_menus,history);
full_filter = fullfile(current_dir,filter);
path_cell = path2cell(current_dir);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(dir_popup,'String',path_cell(end:-1:1),'Value',1)
set(pathbox,'String',current_dir)
set(navlist,'String',filenames,'Value',[])
set(addbut,'Enable','off')
set(openbut,'Enable','off')
end
function showfullpath(varargin)
show_full_path = get(viewfullpath,'Value');
if show_full_path
set(pickslist,'String',full_file_picks)
else
set(pickslist,'String',file_picks)
end
end
function removedupes(varargin)
nodupes = get(remove_dupes,'Value');
if nodupes
num_picks = length(full_file_picks);
[unused,rev_order] = unique(full_file_picks(end:-1:1));
order = sort(num_picks + 1 - rev_order);
full_file_picks = full_file_picks(order);
file_picks = file_picks(order);
if show_full_path
set(pickslist,'String',full_file_picks,'Value',[])
else
set(pickslist,'String',file_picks,'Value',[])
end
set([removebut,moveupbut,movedownbut],'Enable','off')
end
end
function moveup(varargin)
value = get(pickslist,'Value');
set(removebut,'Enable','on')
n = length(file_picks);
omega = 1:n;
index = zeros(1,n);
index(value - 1) = omega(value);
index(setdiff(omega,value - 1)) = omega(setdiff(omega,value));
file_picks = file_picks(index);
full_file_picks = full_file_picks(index);
value = value - 1;
if show_full_path
set(pickslist,'String',full_file_picks,'Value',value)
else
set(pickslist,'String',file_picks,'Value',value)
end
if min(value) == 1
set(moveupbut,'Enable','off')
end
set(movedownbut,'Enable','on')
end
function movedown(varargin)
value = get(pickslist,'Value');
set(removebut,'Enable','on')
n = length(file_picks);
omega = 1:n;
index = zeros(1,n);
index(value + 1) = omega(value);
index(setdiff(omega,value + 1)) = omega(setdiff(omega,value));
file_picks = file_picks(index);
full_file_picks = full_file_picks(index);
value = value + 1;
if show_full_path
set(pickslist,'String',full_file_picks,'Value',value)
else
set(pickslist,'String',file_picks,'Value',value)
end
if max(value) == n
set(movedownbut,'Enable','off')
end
set(moveupbut,'Enable','on')
end
function togglefilter(varargin)
value = get(showallfiles,'Value');
if value
filter = '*';
re_filter = '';
set([filter_ed,refilter_ed],'Enable','off')
else
filter = get(filter_ed,'String');
re_filter = get(refilter_ed,'String');
set([filter_ed,refilter_ed],'Enable','on')
end
full_filter = fullfile(current_dir,filter);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(navlist,'String',filenames,'Value',[])
set(addbut,'Enable','off')
end
function setfilspec(varargin)
filter = get(filter_ed,'String');
if isempty(filter)
filter = '*';
set(filter_ed,'String',filter)
end
% Process file spec if a subdirectory was included.
[p,f,e] = fileparts(filter);
if ~isempty(p)
newpath = fullfile(current_dir,p,'');
set(pathbox,'String',newpath)
filter = [f,e];
if isempty(filter)
filter = '*';
end
set(filter_ed,'String',filter)
change_path();
end
full_filter = fullfile(current_dir,filter);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(navlist,'String',filenames,'Value',[])
set(addbut,'Enable','off')
end
function setrefilter(varargin)
re_filter = get(refilter_ed,'String');
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(navlist,'String',filenames,'Value',[])
set(addbut,'Enable','off')
end
function done(varargin)
% Optional shortcut: click on a file and press 'Done'.
% if isempty(full_file_picks) && strcmp(get(addbut,'Enable'),'on')
% add();
% end
numfiles = length(full_file_picks);
if ~isempty(prop.numfiles)
if numfiles < prop.numfiles(1)
msg = {'Too few files selected.',numstr};
uiwait(errordlg(msg,'','modal'))
return
elseif numfiles > prop.numfiles(end)
msg = {'Too many files selected.',numstr};
uiwait(errordlg(msg,'','modal'))
return
end
end
delete(fig)
end
function cancel(varargin)
prop.output = 'cancel';
delete(fig)
end
function history_cb(varargin)
current_dir = history{varargin{3}};
full_filter = fullfile(current_dir,filter);
path_cell = path2cell(current_dir);
fdir = filtered_dir(full_filter,re_filter);
filenames = {fdir.name}';
filenames = annotate_file_names(filenames,fdir);
set(dir_popup,'String',path_cell(end:-1:1),'Value',1)
set(pathbox,'String',current_dir)
set(navlist,'ListboxTop',1,'Value',[],'String',filenames)
set(addbut,'Enable','off')
set(openbut,'Enable','off')
end
end
% -------------------- Subfunctions --------------------
function c = path2cell(p)
% Turns a path string into a cell array of path elements.
c = strread(p,'%s','delimiter','\\/');
if ispc
c = [{'My Computer'};c];
else
c = [{filesep};c(2:end)];
end
end
function p = cell2path(c)
% Turns a cell array of path elements into a path string.
if ispc
p = fullfile(c{2:end},'');
else
p = fullfile(c{:},'');
end
end
function d = filtered_dir(full_filter,re_filter)
% Like dir, but applies filters and sorting.
p = fileparts(full_filter);
if isempty(p) && full_filter(1) == '/'
p = '/';
end
if exist(full_filter,'dir')
c = cell(0,1);
dfiles = struct('name',c,'date',c,'bytes',c,'isdir',c);
else
dfiles = dir(full_filter);
end
if ~isempty(dfiles)
dfiles([dfiles.isdir]) = [];
end
ddir = dir(p);
ddir = ddir([ddir.isdir]);
% Additional regular expression filter.
if nargin > 1 && ~isempty(re_filter)
if ispc
no_match = cellfun('isempty',regexpi({dfiles.name},re_filter));
else
no_match = cellfun('isempty',regexp({dfiles.name},re_filter));
end
dfiles(no_match) = [];
end
% Set navigator style:
% 1 => mix file and directory names
% 2 => means list all files before all directories
% 3 => means list all directories before all files
% 4 => same as 2 except put . and .. directories first
if isunix
style = 4;
else
style = 4;
end
switch style
case 1
d = [dfiles;ddir];
[unused,index] = sort({d.name});
d = d(index);
case 2
[unused,index1] = sort({dfiles.name});
[unused,index2] = sort({ddir.name});
d = [dfiles(index1);ddir(index2)];
case 3
[unused,index1] = sort({dfiles.name});
[unused,index2] = sort({ddir.name});
d = [ddir(index2);dfiles(index1)];
case 4
[unused,index1] = sort({dfiles.name});
dot1 = find(strcmp({ddir.name},'.'));
dot2 = find(strcmp({ddir.name},'..'));
ddot1 = ddir(dot1);
ddot2 = ddir(dot2);
ddir([dot1,dot2]) = [];
[unused,index2] = sort({ddir.name});
d = [ddot1;ddot2;dfiles(index1);ddir(index2)];
end
end
function drives = getdrives
% Returns a cell array of drive names on Windows.
letters = char('A':'Z');
num_letters = length(letters);
drives = cell(1,num_letters);
for i = 1:num_letters
if exist([letters(i),':\'],'dir');
drives{i} = [letters(i),':'];
end
end
drives(cellfun('isempty',drives)) = [];
end
function filenames = annotate_file_names(filenames,dir_listing)
% Adds a trailing filesep character to directory names.
fs = filesep;
for i = 1:length(filenames)
if dir_listing(i).isdir
filenames{i} = [filenames{i},fs];
end
end
end
function hist_menus = make_history_cm(cb,hist_cm,hist_menus,history)
% Make context menu for history.
if ~isempty(hist_menus)
delete(hist_menus)
end
num_hist = length(history);
hist_menus = zeros(1,num_hist);
for i = 1:num_hist
hist_menus(i) = uimenu(hist_cm,'Label',history{i},...
'Callback',{cb,i});
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
reconstructXsensMotion.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/reconstructXsensMotion.m
| 2,018 |
utf_8
|
2f580dc44896b8d1f48068954ce01bda
|
function recmot=reconstructXsensMotion(Tensor,skel,mot,varargin)
switch nargin
case 3
consideredJoints=mot.animated';
case 4
consideredJoints=varargin{1};
otherwise
error('Wrong number of input arguments!');
end
X0 = cell(1,Tensor.numNaturalModes);
for i=1:Tensor.numNaturalModes
X0{i} = ones(1,Tensor.dimNaturalModes(i))/Tensor.dimNaturalModes(i);
end
fprintf('Computing average accelerations out of Tensor...\n');
[s,m] = constructMotion(Tensor,X0,skel);
m=addAccToMot(m);
fprintf('...done.\n');
mot = changeFrameRate(skel,mot,m.samplingRate);
fprintf('\nFinding subsequence of captured motion...\n');
XsensData = mot.jointAccelerations{consideredJoints};
refData = m.jointAccelerations{consideredJoints};
[xx,start] = cutXsensData(XsensData,refData);
mot=cutMotion(mot,start,start+m.nframes-1);
% fprintf('...done. Start frame: %i, end frame: %i\n',startOS,endOS);
% Warpen der aufgenommenen Bewegung
% fprintf('\nWarping the captured motion...\n');
% [D,w] = pointCloudDTW(m,mot,'a',mot.animated',0);
% mot = warpMotion(w,s,mot);
% fprintf('...done.\n');
fprintf('\nComputing coefficients for reconstructing the captured motion...\n');
set=defaultSet;
set.regardedJoints=consideredJoints;
res = reconstructMotion(Tensor,skel,mot,'set',set);
fprintf('...done.\n');
recmot = res.motRec;
% recmot = warpMotion(fliplr(w),skel,recmot);
end
function [Xsens_cut,start] = cutXsensData(XsensData,refdata)
XsensNorm = normOfColumns(XsensData);
refNorm = normOfColumns(refdata);
mindist = inf;
for i=1:length(XsensNorm)-length(refNorm)+1
dist = sum(abs(XsensNorm(i:i+length(refNorm)-1)-refNorm));
if dist<mindist
mindist=dist;
start=i;
end
end
Xsens_cut = XsensData(:,start:start+length(refNorm)-1);
plot(XsensNorm(:,start:start+length(refNorm)-1));
hold all;
plot(refNorm);
drawnow();
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
linearCombinationOfSkeletons.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/linearCombinationOfSkeletons.m
| 945 |
utf_8
|
e48142f9ab8c6c415213225131adfcd5
|
% example: skel=linearCombinationOfSkeletons({skel1,skel2},[0.5 0.5]);
function skel=linearCombinationOfSkeletons(skels,weights)
numSkeletons=length(skels);
if numSkeletons~=length(weights)
error('Error: Number of skeletons must equal number of weights!');
end
% Initialization
skel=skels{1};
skel.filename='synthesizedSkeleton';
skel.version=[];
skel.name=[];
for i=1:skel.njoints
skel.nodes(i).length = skel.nodes(i).length*weights(1);
skel.nodes(i).direction = skel.nodes(i).direction*weights(1);
for j=2:numSkeletons
skel.nodes(i).length = skel.nodes(i).length+weights(j)*skels{j}.nodes(i).length;
skel.nodes(i).direction = skel.nodes(i).direction+weights(j)*skels{j}.nodes(i).direction;
end
skel.nodes(i).direction = skel.nodes(i).direction/normOfColumns(skel.nodes(i).direction);
skel.nodes(i).offset = skel.nodes(i).direction*skel.nodes(i).length;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
findOptimalPCtransformation.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/findOptimalPCtransformation.m
| 1,655 |
utf_8
|
578de504452a5ed36113682f0e377274
|
function T = findOptimalPCtransformation(pc1,pc2)
% pc1(1,:)=pc1(1,:)*9.81;
% pc1(2,:)=-pc1(2,:);
% pc1(3,:)=pc1(3,:)/9.81;
T.pc1 = pc1;
T.pc2 = pc2;
T.pc1_abs = normOfColumns(T.pc1);
T.pc2_abs = normOfColumns(T.pc2);
figure()
plot(T.pc1_abs);
hold all
plot(T.pc2_abs);
hold off
drawnow;
figure()
plot(T.pc1');
hold all
plot(T.pc2');
hold off
drawnow;
% startValue = [0 0 0 1];
% lb = [-2*pi -2*pi -2*pi 0];
% ub = [ 2*pi 2*pi 2*pi 2];
startValue = [0 0 0];
lb = [-2*pi -2*pi -2*pi];
ub = [ 2*pi 2*pi 2*pi];
options = optimset( 'Display','iter' ,...
'MaxFunEvals',100000,...
'MaxIter',100,...
'TolFun',0.001);%,...
% 'PlotFcns', @optimplotx);
X = lsqnonlin(@(x) objfunLocal(x,pc1,pc2),startValue,lb,ub,options);
T.rotation = X(1:3);
% T.scale = X(4:6);
T.qR = C_euler2quat(X(1:3)');
% T.pc1_new = C_quatrot(pc1,T.qR) .* repmat(T.scale',1,size(T.pc1,2));
T.pc1_new = C_quatrot(pc1,T.qR);
figure
subplot(3,1,1);plot(T.pc1')
subplot(3,1,2);plot(T.pc2')
subplot(3,1,3);plot(T.pc1_new')
drawnow();
figure()
plot(T.pc1_new');
hold all
plot(T.pc2');
hold off
drawnow;
end
%% local functions
function f = objfunLocal(x,pc1,pc2)
% qx = rotquat(x(1),'x');
% qy = rotquat(x(2),'y');
% qz = rotquat(x(3),'z');
% q = C_quatmult(qz,C_quatmult(qy,qx));
% pc1(1,:)=pc1(1,:)*x(4);
% pc1(2,:)=pc1(1,:)*x(5);
% pc1(3,:)=pc1(1,:)*x(6);
q = C_euler2quat(x(1:3)');
% f = C_quatrot(pc1,q) * x(4) - pc2;
f = C_quatrot(pc1,q) - pc2;
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
resampleMot.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/resampleMot.m
| 3,898 |
utf_8
|
c02cbf66d8145c49bea30ef2d6ce0c30
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% function resampleMot
% resamples a motion to create homogeneous point cloud trajectories
%
% resmot = resampleMot(skel,mot[,joints[,numberOfSamples]])
%
% input:
% - skel: skeleton (required for forward kinematics)
% - mot: motion to be resampled
% - joints: joint IDs of regarded joints
% - numberOfSamples: new number of samples (== resmot.nframes)
%
% output:
% - resmot: resampled motion
%
% original motion:
% oo-o--o----o------o-----------o----------------o------------------------o
% resampled motion:
% x--------x--------x--------x--------x--------x--------x--------x--------x
% o = old samples, x = new samples
%
% author: Jochen Tautges ([email protected])
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function resmot = resampleMot(skel,mot,varargin)
rootTranslation = mot.rootTranslation;
mot.rootTranslation(:,:) = 0;
mot.jointTrajectories = C_forwardKinematicsQuat(skel,mot);
switch nargin
case 2
numberOfSamples = mot.nframes;
joints = 1:mot.njoints;
case 3
numberOfSamples = varargin{1};
joints = 1:mot.njoints;
case 4
numberOfSamples = varargin{1};
joints = varargin{2};
otherwise
help resampleMot;
error('Wrong number of argins!');
end
newRootTranslation = zeros(3,numberOfSamples);
newRootTranslation(:,1) = rootTranslation(:,1);
computeAcc = false;
if isfield(mot,'jointAccelerations')
computeAcc = true;
accs = cell2mat(mot.jointAccelerations);
accs_new = zeros(size(accs,1),numberOfSamples);
accs_new(:,1) = accs(:,1);
end
computeVel = false;
if isfield(mot,'jointVelocities')
computeVel = true;
vels = cell2mat(mot.jointVelocities);
vels_new = zeros(size(vels,1),numberOfSamples);
vels_new(:,1) = vels(:,1);
end
pos = cell2mat(mot.jointTrajectories(joints));
quats = cell2mat(mot.rotationQuat(mot.animated));
dists = sqrt(sum(diff(pos,1,2).^2));
delta = sum(dists)/(numberOfSamples-1);
% delta = mean(dists);
quats_new = zeros(size(quats,1),numberOfSamples);
quats_new(:,1) = quats(:,1);
akk = 0;
tmp = 0;
i=1;
counter=1;
while i<mot.nframes
step = dists(i)-tmp;
if akk + step < delta
tmp = 0;
akk = akk + step;
i = i+1;
else
counter = counter+1;
tmp = tmp + delta - akk;
alpha = tmp / dists(i);
quats_new(:,counter) = alpha * quats(:,i+1) + (1-alpha) * quats(:,i);
if computeAcc
accs_new(:,counter) = alpha * accs(:,i+1) + (1-alpha) * accs(:,i);
end
if computeVel
vels_new(:,counter) = alpha * vels(:,i+1) + (1-alpha) * vels(:,i);
end
newRootTranslation(:,counter) = alpha * rootTranslation(:,i+1) + (1-alpha) * rootTranslation(:,i);
akk = 0;
end
end
dofs = getDOFsFromSkel(skel);
resmot = emptyMotion(mot);
resmot.nframes = numberOfSamples;
resmot.rootTranslation = newRootTranslation(:,1:numberOfSamples);
resmot.rotationQuat = mat2cell(quats_new(:,1:numberOfSamples),dofs.quat,numberOfSamples);
resmot.jointTrajectories = C_forwardKinematicsQuat(skel,resmot);
if computeAcc
resmot.jointAccelerations = mat2cell(accs_new(:,1:numberOfSamples),dofs.pos,resmot.nframes);
end
if computeVel
resmot.jointVelocities = mat2cell(vels_new(:,1:numberOfSamples),dofs.pos,resmot.nframes);
end
resmot.boundingBox = computeBoundingBox(resmot);
% if ~isempty(mot.rotationEuler)
% resmot = C_convert2euler(skel,resmot);
% end
mot.documentation = 'resampled';
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
recMotFromSimWiiData.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/Optimization/recMotFromSimWiiData.m
| 3,273 |
utf_8
|
19bb2e00e269f2cb7d9d4d6ed272b114
|
function res = recMotFromSimWiiData(ttQuats,ttRootPos,skel,mot,joints)
mot = fitMotion(skel,mot);
refMot = extractMotFromTTensor(skel,ttQuats,ttRootPos,[1,1,1]);
[D,w] = pointCloudDTW_pos(refMot,mot,2);
mot = warpMotion(w,skel,mot);
res.skel = skel;
res.origmot = mot;
res.joints = joints;
optimStruct.TensorInfo.jointModeID = 3;
optimStruct.TensorInfo.frameModeID = 2;
optimStruct.TensorInfo.dofModeID = 1;
optimStruct.TensorInfo.modesForOpt = [4,5,6];
optimStruct.ttQuatsFact = tucker_als(ttQuats,ttQuats.size);
optimStruct.ttRootPosFact = tucker_als(ttRootPos,ttRootPos.size);
% simulate local accelerations
res.noise = 1;
optimStruct.simWiiData = cell(mot.njoints,1);
for j=joints
optimStruct.simWiiData{j} = simulateLocalAccs(skel,mot,j,res.noise);
end
% ttQuats_prepared = ttm(ttQuats_factorized.core,...
% ttQuats_factorized.U(1:Tensor.numTechnicalModes),...
% 1:Tensor.numTechnicalModes);
% ttRootPos_prepared = ttm(ttRootPos_factorized.core,...
% ttRootPos_factorized.U(1:Tensor.numTechnicalModes-1),...
% 1:Tensor.numTechnicalModes-1);
optimStruct.skel = skel;
optimStruct.origmot = res.origmot;
optimStruct.joints = joints;
optimStruct.ttQuats = ttQuats;
optimStruct.ttRootPos = ttRootPos;
% optimStruct.ttQuats_prepared = ttQuats_prepared;
% optimStruct.ttRootPos_prepared = ttRootPos_prepared;
% optimization options ----------------------------------------------------
options = optimset( 'Display','iter',...
'MaxFunEvals',100000,...
'MaxIter',100,...
'TolFun',0.001,...
'PlotFcns', @optimplotx);
optimStruct.dimQuats = ttQuats.size;
startValue = buildStartValue(optimStruct.dimQuats(optimStruct.TensorInfo.modesForOpt));
lb = -0.2 * ones(1,length(startValue));
ub = 1.2 * ones(1,length(startValue));
% -------------------------------------------------------------------------
coeffs = lsqnonlin(@(x) objfunWii(x,optimStruct),startValue,lb,ub,options);
res.coeffs = mat2cell(coeffs,1,optimStruct.dimQuats(optimStruct.TensorInfo.modesForOpt))';
res.recmot = ttensor2mot(optimStruct.ttQuatsFact,optimStruct.ttRootPosFact,optimStruct.skel,res.coeffs,optimStruct.TensorInfo,optimStruct.origmot);
res.recmot.boundingBox = computeBoundingBox(res.recmot);
end
%%
function f = objfunWii(x,optimStruct)
X = mat2cell(x,1,optimStruct.dimQuats(optimStruct.TensorInfo.modesForOpt))';
motRec = ttensor2mot(optimStruct.ttQuatsFact,optimStruct.ttRootPosFact,optimStruct.skel,X,optimStruct.TensorInfo,optimStruct.origmot);
motRec = addAccToMot(motRec);
motRec = computeLocalSystems(optimStruct.skel,motRec);
f= [];
for i=optimStruct.joints
recData = motRec.jointAccelerations{i};
recData(2,:)= recData(2,:)+9.81;
P = C_quatrot(recData,C_quatinv(motRec.localSystems{i}));
Q = optimStruct.simWiiData{i};
T = findOptimalPCtransformation(P,Q);
f = [f;T.pc1_new - Q];
end
end
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
dtwModStep.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/synthesis/dtwModStep.m
| 2,200 |
utf_8
|
e95ff16ef5caf73be712e049bcd19e07
|
function [Dist,D,k,w,d]=dtwModStep(t1,r1)
%Dynamic Time Warping Algorithm
%Dist is unnormalized distance between t1 and r1
%D is the accumulated distance matrix (GDM)
%k is the normalizing factor
%w is the optimal path
%d is the local distance matrix (LDM)
%t1 is the vector you are testing against
%r1 is the vector you are testing
%Transpose the given Matrices
t1=t1';
r1=r1';
[rows,N]=size(t1);
[rows,M]=size(r1);
d=zeros(N,M);
%Calculation of the LDM:
for m=1:M
for n=1:N
%This calculates the distance per frame only for a 1-dim signal:
%d(n,m)=eukl_dist1d(t1(n),r1(m));
%You should replace this by your own distance measure for n-dim signals:
d(n,m)=highDimDistance(t1(:,n),r1(:,m),rows);
% d(n,m)=winDist(t1(:,n-9:n+9),r1(:,m-9:m+9));
end
end
%Calculation of the GDM:
D=NaN(size(d));
D(1,1)=d(1,1);
for n=2:N
D(n,1:2)=d(n,1:2)+D(n-1,1:2);
end
for m=2:M
D(1:2,m)=d(1:2,m)+D(1:2,m-1);
end
for n=3:N
for m=3:M
D(n,m)=d(n,m)+min([D(n-2,m-1),D(n-1,m-1),D(n-1,m-2)]);
end
end
%Search of the optimal path on the given matrix:
Dist=D(N,M);
n=N;
m=M;
k=1;
w=[];
w(1,:)=[N,M];
while ((n+m)~=2)
if (n-1)==0
m=m-1;
elseif (m-1)==0
n=n-1;
else
[values,number]=min([D(n-1,m-1),D(n-1,m-2),D(n-1,m-2)]);
switch number
case 2
n=n-1;m=m-2;
case 3
m=m-2;n=n-1;
case 1
n=n-1;
m=m-1;
end
end
k=k+1;
w=cat(1,w,[n,m]);
end
%End of DTW Algorithm
% Euclidian distance measure
function [distance]=eukl_dist1d(a1,b1)
distance=abs((a1-b1));
% Add your function for the distance measurement here:
function [distance]=highDimDistance(a1,b1,row)
sum=0;
%Berechnung des n-Dim. euklidischen Abstandes
for i=1:row
sum=sum+(a1(i)-b1(i))^2;
end
distance=sqrt(sum);
function [dist]=dotDist(a,b)
dist=1-dot(a/sqrt(sum(a.*a)),b/sqrt(sum(b.*b)));
function [dist]=winDist(a,b)
a=sum(a,2);
b=sum(b,2);
dist=1-dot(a/sqrt(sum(a.*a)),b/sqrt(sum(b.*b)));
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
dtw.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/synthesis/dtw.m
| 2,597 |
utf_8
|
a29941e18ad22ac3aedb4694fc9168bf
|
function [Dist,D,k,w,d]=dtw(varargin)
%Dynamic Time Warping Algorithm
%Dist is unnormalized distance between t1 and r1
%D is the accumulated distance matrix (GDM)
%k is the normalizing factor
%w is the optimal path
%d is the local distance matrix (LDM)
%t1 is the vector you are testing against
%r1 is the vector you are testing
switch nargin
case 1
d = varargin{1};
N = size(d,1);
M = size(d,2);
case 2
%Transpose the given Matrices
t1 = varargin{1}';
r1 = varargin{2}';
[rows,N] = size(t1);
[rows,M] = size(r1);
d = zeros(N,M);
%Calculation of the LDM:
for m=1:M
for n=1:N
%This calculates the distance per frame only for a 1-dim signal:
%d(n,m)=eukl_dist1d(t1(n),r1(m));
%You should replace this by your own distance measure for n-dim signals:
d(n,m)=highDimDistance(t1(:,n),r1(:,m));
% d(n,m)=winDist(t1(:,n-9:n+9),r1(:,m-9:m+9));
end
end
otherwise
error('Wrong num of args!\n');
end
%Calculation of the GDM:
D=inf(size(d));
D(1,1)=d(1,1);
for n=2:N
D(n,1)=d(n,1)+D(n-1,1);
% D(n,2)=d(n,2)+D(n-1,2);
end
for m=2:M
D(1,m)=d(1,m)+D(1,m-1);
% D(2,m)=d(2,m)+D(2,m-1);
end
for n=2:N
for m=2:M
D(n,m)=d(n,m)+min([D(n-1,m-1),D(n-1,m),D(n,m-1)]);
% D(n,m)=d(n,m)+min([D(n-1,m-2),D(n-1,m-1),D(n-2,m-1)]);
end
end
%Search of the optimal path on the given matrix:
Dist=D(N,M);
n=N;
m=M;
k=1;
w=[];
w(1,:)=[N,M];
while ((n+m)~=2)
if (n-1)==0
m=m-1;
elseif (m-1)==0
n=n-1;
else
[values,number]=min([D(n-1,m-1),D(n-1,m),D(n,m-1)]);
switch number
case 2
n=n-1;
case 3
m=m-1;
case 1
n=n-1;
m=m-1;
end
end
k=k+1;
w=cat(1,w,[n,m]);
end
%End of DTW Algorithm
% Euclidian distance measure
function [distance]=eukl_dist1d(a1,b1)
distance=abs((a1-b1));
% Add your function for the distance measurement here:
function [distance]=highDimDistance(a1,b1)
% sum=0;
tmp=a1-b1;
distance=sqrt(sum(tmp.^2));
%Berechnung des n-Dim. euklidischen Abstandes
function [dist]=dotDist(a,b)
dist=1-dot(a/sqrt(sum(a.*a)),b/sqrt(sum(b.*b)));
function [dist]=winDist(a,b)
a=sum(a,2);
b=sum(b,2);
dist=1-dot(a/sqrt(sum(a.*a)),b/sqrt(sum(b.*b)));
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
ldm.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/synthesis/ldm.m
| 1,078 |
utf_8
|
c37961027af9710fa4d03ca8ab9e3f7b
|
function [d]=ldm(t1,r1)
%Dynamic Time Warping Algorithm
%d is the local distance matrix (LDM)
%Transpose the given Matrices
t1=t1';
r1=r1';
[rows,N]=size(t1);
[rows,M]=size(r1);
d=zeros(N,M);
%Calculation of the LDM:
fprintf('Row: ');
tic;
for m=1:M
for n=1:N
%This calculates the distance per frame only for a 1-dim signal:
%d(n,m)=eukl_dist1d(t1(n),r1(m));
%You should replace this by your own distance measure for n-dim signals:
d(n,m)=highDimDistance(t1(:,n),r1(:,m),rows);
end
fprintf('\b\b\b\b');
fprintf('%4i',m);
end
time=toc;
disp(['Calculated LDM in:' num2str(time) ' seconds']);
% Euclidian distance measure
function [distance]=eukl_dist1d(a1,b1)
distance=abs((a1-b1));
% Add your function for the distance measurement here:
function [distance]=highDimDistance(a1,b1,row)
%Berechnung des n-Dim. euklidischen Abstandes
distance=a1-b1;
distance=distance'*distance;
% for i=1:row
% sum=sum+(a1(i)-b1(i))^2;
% end
% distance=sum;
|
github
|
umariqb/3D_Pose_Estimation_CVPR2016-master
|
grMinSpanTree.m
|
.m
|
3D_Pose_Estimation_CVPR2016-master/tools_intern/synthesis/GrTheory/grMinSpanTree.m
| 2,174 |
utf_8
|
de38e2d0ce1875aae138ca0a0ca15931
|
function nMST=grMinSpanTree(E)
% Function nMST=grMinSpanTree(E) solve
% the minimal spanning tree problem for a connected graph.
% Input parameter:
% E(m,2) or (m,3) - the edges of graph and their weight;
% 1st and 2nd elements of each row is numbers of vertexes;
% 3rd elements of each row is weight of edge;
% m - number of edges.
% If we set the array E(m,2), then all weights is 1.
% Output parameter:
% nMST(n-1,1) - the list of the numbers of edges included
% in the minimal (weighted) spanning tree in the including order.
% Uses the greedy algorithm.
% Author: Sergiy Iglin
% e-mail: [email protected]
% personal page: http://iglin.exponenta.ru
% ============= Input data validation ==================
if nargin<1,
error('There are no input data!')
end
[m,n,E] = grValidation(E); % E data validation
% ============= The data preparation ==================
En=[(1:m)',E]; % we add the numbers
En(:,2:3)=sort(En(:,2:3)')'; % edges on increase order
ln=find(En(:,2)==En(:,3)); % the loops numbers
En=En(setdiff([1:size(En,1)]',ln),:); % we delete the loops
[w,iw]=sort(En(:,4)); % sort by weight
Ens=En(iw,:); % sorted edges
% === We build the minimal spanning tree by the greedy algorithm ===
Emst=Ens(1,:); % 1st edge include to minimal spanning tree
Ens=Ens(2:end,:); % rested edges
while (size(Emst,1)<n-1)&(~isempty(Ens)),
Emst=[Emst;Ens(1,:)]; % we add next edge to spanning tree
Ens=Ens(2:end,:); % rested edges
if any((Emst(end,2)==Emst(1:end-1,2))&...
(Emst(end,3)==Emst(1:end-1,3))) | ...
IsCycle(Emst(:,2:3)), % the multiple edge or cycle
Emst=Emst(1:end-1,:); % we delete the last added edge
end
end
nMST=Emst(:,1); % numbers of edges
return
function ic=IsCycle(E); % true, if graph E have cycle
n=max(max(E)); % number of vertexes
A=zeros(n);
A((E(:,1)-1)*n+E(:,2))=1;
A=A+A'; % the connectivity matrix
p=sum(A); % the vertexes power
ic=false;
while any(p<=1), % we delete all tails
nc=find(p>1); % rested vertexes
if isempty(nc),
return
end
A=A(nc,nc); % new connectivity matrix
p=sum(A); % new powers
end
ic=true;
return
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.