File size: 11,364 Bytes
861e32a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 |
'''ResNet in PyTorch.
For Pre-activation ResNet, see 'preact_resnet.py'.
Reference:
[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
Deep Residual Learning for Image Recognition. arXiv:1512.03385
'''
import torch
import torch.nn as nn
import torch.nn.functional as F
from utils.utils import load_weights
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, in_planes, planes, stride=1):
super().__init__()
self.conv1 = nn.Conv2d(
in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3,
stride=1, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(self.expansion*planes)
)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = self.bn2(self.conv2(out))
out += self.shortcut(x)
out = F.relu(out)
return out
class Bottleneck(nn.Module):
expansion = 2
def __init__(self, in_planes, planes, stride=1):
super().__init__()
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3,
stride=stride, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes, self.expansion *
planes, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(self.expansion*planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(self.expansion*planes)
)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = F.relu(self.bn2(self.conv2(out)))
out = self.bn3(self.conv3(out))
out += self.shortcut(x)
out = F.relu(out)
return out
class ResNet(nn.Module):
def __init__(self, block, num_blocks, channels=4, num_classes=10, gap_output=False, before_gap_output=False, visualize=False):
super().__init__()
self.block = block
self.num_blocks = num_blocks
self.in_planes = 64
self.gap_output = gap_output
self.before_gap_out = before_gap_output
self.visualize = visualize
self.conv1 = nn.Conv2d(channels, 64, kernel_size=3,
stride=1, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2)
self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2)
self.layer5 = None
self.layer6 = None
if not gap_output and not before_gap_output:
self.linear = nn.Linear(512*block.expansion, num_classes)
def add_top_blocks(self, num_classes=1):
self.layer5 = self._make_layer(Bottleneck, 512, 2, stride=2)
self.layer6 = self._make_layer(Bottleneck, 512, 2, stride=2)
if not self.gap_output and not self.before_gap_out:
self.linear = nn.Linear(1024, num_classes)
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.in_planes, planes, stride))
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = self.layer1(out)
out = self.layer2(out)
out = self.layer3(out)
out4 = self.layer4(out)
if self.before_gap_out:
return out4
if self.layer5:
out5 = self.layer5(out4)
out6 = self.layer6(out5)
n, c, _, _ = out6.size()
out = out6.view(n, c, -1).mean(-1)
if self.gap_output:
return out
out = self.linear(out)
if self.visualize:
return out, out4, out6
return out
class Encoder(nn.Module):
def __init__(self, channels):
super().__init__()
self.in_planes = 64
self.conv1 = nn.Conv2d(channels, 64, kernel_size=3,
stride=1, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.layer1 = self._make_layer(BasicBlock, 64, 2, stride=1)
self.layer2 = self._make_layer(BasicBlock, 128, 2, stride=2)
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.in_planes, planes, stride))
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = self.layer1(out)
out = self.layer2(out)
return out
class SharedBottleneck(nn.Module):
def __init__(self, in_planes):
super().__init__()
self.in_planes = in_planes
self.layer3 = self._make_layer(BasicBlock, 256, 2, stride=2)
self.layer4 = self._make_layer(BasicBlock, 512, 2, stride=2)
self.layer5 = self._make_layer(Bottleneck, 512, 2, stride=2)
self.layer6 = self._make_layer(Bottleneck, 512, 2, stride=2)
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.in_planes, planes, stride))
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x):
out = self.layer3(x)
out = self.layer4(out)
out = self.layer5(out)
out = self.layer6(out)
n, c, _, _ = out.size()
out = out.view(n, c, -1).mean(-1)
return out
class Classifier(nn.Module):
def __init__(self, num_classes, in_planes=512, visualize=False):
super().__init__()
self.in_planes = in_planes
self.visualize = visualize
self.layer5 = self._make_layer(Bottleneck, 512, 2, stride=2)
self.layer6 = self._make_layer(Bottleneck, 512, 2, stride=2)
self.linear = nn.Linear(1024, num_classes)
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.in_planes, planes, stride))
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x):
out = self.layer5(x)
feature_maps = self.layer6(out)
n, c, _, _ = feature_maps.size()
out = feature_maps.view(n, c, -1).mean(-1)
out = self.linear(out)
if self.visualize:
return out, feature_maps
return out
class SBOnet(nn.Module):
"""SBOnet.
Parameters:
- shared: True to share the Bottleneck between the two sides, False for the 'concat' version.
- weights: path to pretrained weights of patch classifier for Encoder branches
"""
def __init__(self, shared=True, num_classes=1, weights=None):
super().__init__()
self.shared = shared
self.encoder_sx = Encoder(channels=2)
self.encoder_dx = Encoder(channels=2)
self.shared_resnet = SharedBottleneck(in_planes=128 if shared else 256)
if weights:
load_weights(self.encoder_sx, weights)
load_weights(self.encoder_dx, weights)
self.classifier_sx = nn.Linear(1024, num_classes)
self.classifier_dx = nn.Linear(1024, num_classes)
def forward(self, x):
x_sx, x_dx = x
# Apply Encoder
out_sx = self.encoder_sx(x_sx)
out_dx = self.encoder_dx(x_dx)
# Shared layers
if self.shared:
out_sx = self.shared_resnet(out_sx)
out_dx = self.shared_resnet(out_dx)
out_sx = self.classifier_sx(out_sx)
out_dx = self.classifier_dx(out_dx)
else: # Concat version
out = torch.cat([out_sx, out_dx], dim=1)
out = self.shared_resnet(out)
out_sx = self.classifier_sx(out)
out_dx = self.classifier_dx(out)
out = torch.cat([out_sx, out_dx], dim=0)
return out
class SEnet(nn.Module):
"""SEnet.
Parameters:
- weights: path to pretrained weights of patch classifier for PHCResNet18 encoder or path to whole-image classifier
- patch_weights: True if the weights correspond to patch classifier, False if they are whole-image.
In the latter case also Classifier branches will be initialized.
"""
def __init__(self, num_classes=1, weights=None, patch_weights=True, visualize=False):
super().__init__()
self.visualize = visualize
self.resnet18 = ResNet18(
num_classes=num_classes, channels=2, before_gap_output=True)
if weights:
print('Loading weights for resnet18 from ', weights)
load_weights(self.resnet18, weights)
self.classifier_sx = Classifier(num_classes, visualize=visualize)
self.classifier_dx = Classifier(num_classes, visualize=visualize)
if not patch_weights and weights:
print('Loading weights for classifiers from ', weights)
load_weights(self.classifier_sx, weights)
load_weights(self.classifier_dx, weights)
def forward(self, x):
x_sx, x_dx = x
# Apply Encoder
out_enc_sx = self.resnet18(x_sx)
out_enc_dx = self.resnet18(x_dx)
if self.visualize:
out_sx, act_sx = self.classifier_sx(out_enc_sx)
out_dx, act_dx = self.classifier_dx(out_enc_dx)
else:
# Apply refiner blocks + classifier
out_sx = self.classifier_sx(out_enc_sx)
out_dx = self.classifier_dx(out_enc_dx)
out = torch.cat([out_sx, out_dx], dim=0)
if self.visualize:
return out, out_enc_sx, out_enc_dx, act_sx, act_dx
return out
def ResNet18(num_classes=10, channels=4, gap_output=False, before_gap_output=False, visualize=False):
return ResNet(BasicBlock,
[2, 2, 2, 2],
num_classes=num_classes,
channels=channels,
gap_output=gap_output,
before_gap_output=before_gap_output,
visualize=visualize)
def ResNet50(num_classes=10, channels=4):
return ResNet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes, channels=channels)
|