Spaces:
Running
on
T4
Running
on
T4
File size: 5,928 Bytes
4562a06 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 |
import torch
import torch.nn as nn
from diffusionsfm.model.dit import TimestepEmbedder
import ipdb
def modulate(x, shift, scale):
return x * (1 + scale.unsqueeze(-1).unsqueeze(-1)) + shift.unsqueeze(-1).unsqueeze(
-1
)
def _make_fusion_block(features, use_bn, use_ln, dpt_time, resolution):
return FeatureFusionBlock_custom(
features,
nn.ReLU(False),
deconv=False,
bn=use_bn,
expand=False,
align_corners=True,
dpt_time=dpt_time,
ln=use_ln,
resolution=resolution
)
def _make_scratch(in_shape, out_shape, groups=1, expand=False):
scratch = nn.Module()
out_shape1 = out_shape
out_shape2 = out_shape
out_shape3 = out_shape
out_shape4 = out_shape
if expand == True:
out_shape1 = out_shape
out_shape2 = out_shape * 2
out_shape3 = out_shape * 4
out_shape4 = out_shape * 8
scratch.layer1_rn = nn.Conv2d(
in_shape[0],
out_shape1,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=groups,
)
scratch.layer2_rn = nn.Conv2d(
in_shape[1],
out_shape2,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=groups,
)
scratch.layer3_rn = nn.Conv2d(
in_shape[2],
out_shape3,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=groups,
)
scratch.layer4_rn = nn.Conv2d(
in_shape[3],
out_shape4,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=groups,
)
return scratch
class ResidualConvUnit_custom(nn.Module):
"""Residual convolution module."""
def __init__(self, features, activation, bn, ln, dpt_time=False, resolution=16):
"""Init.
Args:
features (int): number of features
"""
super().__init__()
self.bn = bn
self.ln = ln
self.groups = 1
self.conv1 = nn.Conv2d(
features,
features,
kernel_size=3,
stride=1,
padding=1,
bias=not self.bn,
groups=self.groups,
)
self.conv2 = nn.Conv2d(
features,
features,
kernel_size=3,
stride=1,
padding=1,
bias=not self.bn,
groups=self.groups,
)
nn.init.kaiming_uniform_(self.conv1.weight)
nn.init.kaiming_uniform_(self.conv2.weight)
if self.bn == True:
self.bn1 = nn.BatchNorm2d(features)
self.bn2 = nn.BatchNorm2d(features)
if self.ln == True:
self.bn1 = nn.LayerNorm((features, resolution, resolution))
self.bn2 = nn.LayerNorm((features, resolution, resolution))
self.activation = activation
if dpt_time:
self.t_embedder = TimestepEmbedder(hidden_size=features)
self.adaLN_modulation = nn.Sequential(
nn.SiLU(), nn.Linear(features, 3 * features, bias=True)
)
def forward(self, x, t=None):
"""Forward pass.
Args:
x (tensor): input
Returns:
tensor: output
"""
if t is not None:
# Embed timestamp & calculate shift parameters
t = self.t_embedder(t) # (B*N)
shift, scale, gate = self.adaLN_modulation(t).chunk(3, dim=1) # (B * N, T)
# Shift & scale x
x = modulate(x, shift, scale) # (B * N, T, H, W)
out = self.activation(x)
out = self.conv1(out)
if self.bn or self.ln:
out = self.bn1(out)
out = self.activation(out)
out = self.conv2(out)
if self.bn or self.ln:
out = self.bn2(out)
if self.groups > 1:
out = self.conv_merge(out)
if t is not None:
out = gate.unsqueeze(-1).unsqueeze(-1) * out
return out + x
class FeatureFusionBlock_custom(nn.Module):
"""Feature fusion block."""
def __init__(
self,
features,
activation,
deconv=False,
bn=False,
ln=False,
expand=False,
align_corners=True,
dpt_time=False,
resolution=16,
):
"""Init.
Args:
features (int): number of features
"""
super(FeatureFusionBlock_custom, self).__init__()
self.deconv = deconv
self.align_corners = align_corners
self.groups = 1
self.expand = expand
out_features = features
if self.expand == True:
out_features = features // 2
self.out_conv = nn.Conv2d(
features,
out_features,
kernel_size=1,
stride=1,
padding=0,
bias=True,
groups=1,
)
nn.init.kaiming_uniform_(self.out_conv.weight)
# The second block sees time
self.resConfUnit1 = ResidualConvUnit_custom(
features, activation, bn=bn, ln=ln, dpt_time=False, resolution=resolution
)
self.resConfUnit2 = ResidualConvUnit_custom(
features, activation, bn=bn, ln=ln, dpt_time=dpt_time, resolution=resolution
)
def forward(self, input, activation=None, t=None):
"""Forward pass.
Returns:
tensor: output
"""
output = input
if activation is not None:
res = self.resConfUnit1(activation)
output += res
output = self.resConfUnit2(output, t)
output = torch.nn.functional.interpolate(
output.float(),
scale_factor=2,
mode="bilinear",
align_corners=self.align_corners,
)
output = self.out_conv(output)
return output
|