Alignment-Lab-AI's picture
Upload folder using huggingface_hub
d5bfab8 verified
import numpy as np
def Process(img, num_colors=3, **param):
"""Converts a 2d image into to 3d image
where each band corresponds to one colour.
Note: Background is not considered a color.
"""
n, m = img.shape
img_cpy = np.zeros((n, m, num_colors))
for color in range(0, num_colors):
img_cpy[:, :, color] = (img == (color+1))*1
return img_cpy
def Change_Colour(img, rule, *args):
"""
The rule is a mapping {0,1}^k -> {0,1,2,...k}
Implemented as n x (k+1) array where $k$ is the number of colors.
-> row 'i' corresponds to rule 'i'.
-> Each rule will be of the form <0,1,0,1...,j>
-> The last entry denotes the color to assign based on first 'k'
entries.
"""
if rule is None:
m, n, no_colors = img.shape
out_img = np.zeros((m, n), dtype=np.int32)
for col in range(no_colors):
out_img[img[:, :, col] == 1] = col + 1
return out_img
def func(arr, rule):
ind = np.where(np.all(rule[:, :-1] == arr.reshape((1, -1)), axis=-1))[0]
if len(ind) == 0:
return 0
elif len(ind) > 1:
raise Exception("More than two Color_Change rules match the input")
else:
return rule[ind[0], -1]
img = np.apply_along_axis(func, 2, img, rule)
return img