-
Notifications
You must be signed in to change notification settings - Fork 3
/
resnet.py
437 lines (381 loc) · 17.9 KB
/
resnet.py
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
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
import torch.nn.functional as F
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
import torch
__all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101',
'resnet152']
model_urls = {
'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
}
def conv3x3(in_planes, out_planes, stride=1, dilation=1):
"""3x3 convolution with padding"""
# original padding is 1; original dilation is 1
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
padding=dilation, bias=False, dilation=dilation)
class Sequential(nn.Sequential):
# def __init__(self, *args):
# super(Sequential, self).__init__()
def forward(self, input, modal=0):
for module in self:
res = module(input, modal)
input = res
return res
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None, dilation=1):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride, dilation)
self.bn1 = nn.BatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.BatchNorm2d(planes)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, dilation=1, isshape=False, modalbn=1):
super(Bottleneck, self).__init__()
self.isshape = isshape
self.modalbn = modalbn
assert modalbn == 1 or modalbn == 2 or modalbn == 3
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
if isshape:
self.bn1_shape = nn.BatchNorm2d(planes)
if modalbn == 2:
self.bn1_ir = nn.BatchNorm2d(planes)
if modalbn == 3:
self.bn1_ir = nn.BatchNorm2d(planes)
self.bn1_modalx = nn.BatchNorm2d(planes)
# original padding is 1; original dilation is 1
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=dilation, bias=False, dilation=dilation)
self.bn2 = nn.BatchNorm2d(planes)
if isshape:
self.bn2_shape = nn.BatchNorm2d(planes)
if modalbn == 2:
self.bn2_ir = nn.BatchNorm2d(planes)
if modalbn == 3:
self.bn2_ir = nn.BatchNorm2d(planes)
self.bn2_modalx = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(planes * 4)
if isshape:
self.bn3_shape = nn.BatchNorm2d(planes * 4)
if modalbn == 2:
self.bn3_ir = nn.BatchNorm2d(planes * 4)
if modalbn == 3:
self.bn3_ir = nn.BatchNorm2d(planes * 4)
self.bn3_modalx = nn.BatchNorm2d(planes * 4)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
if downsample is not None:
if isshape:
self.dsbn_shape = nn.BatchNorm2d(downsample[1].weight.shape[0])
if modalbn == 2:
self.dsbn_ir = nn.BatchNorm2d(downsample[1].weight.shape[0])
if modalbn == 3:
self.dsbn_ir = nn.BatchNorm2d(downsample[1].weight.shape[0])
self.dsbn_modalx = nn.BatchNorm2d(downsample[1].weight.shape[0])
self.stride = stride
def forward(self, x, modal=0):
if modal == 0: # RGB
bbn1 = self.bn1
bbn2 = self.bn2
bbn3 = self.bn3
if self.downsample is not None:
dsbn = self.downsample[1]
elif modal == 1: # IR
bbn1 = self.bn1_ir
bbn2 = self.bn2_ir
bbn3 = self.bn3_ir
if self.downsample is not None:
dsbn = self.dsbn_ir
elif modal == 2: # modalx
bbn1 = self.bn1_modalx
bbn2 = self.bn2_modalx
bbn3 = self.bn3_modalx
if self.downsample is not None:
dsbn = self.dsbn_modalx
elif modal == 3: # shape
assert self.isshape == True
bbn1 = self.bn1_shape
bbn2 = self.bn2_shape
bbn3 = self.bn3_shape
if self.downsample is not None:
dsbn = self.dsbn_shape
residual = x
out = self.conv1(x)
out = bbn1(out)
out = self.relu(out)
out = self.conv2(out)
out = bbn2(out)
out = self.relu(out)
out = self.conv3(out)
out = bbn3(out)
if self.downsample is not None:
residual = dsbn(self.downsample[0](x))
out += residual
outt = F.relu(out)
return outt
class ResNet(nn.Module):
def __init__(self, block, layers, last_conv_stride=2, last_conv_dilation=1, isshape=False, modalbn=1, onlyshallow=False):
self.isshape = isshape
self.modalbn = modalbn
self.inplanes = 64
super(ResNet, self).__init__()
# if onlyshallow:
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
if self.isshape:
self.bn1_shape = nn.BatchNorm2d(64)
if self.modalbn == 2:
self.bn1_ir = nn.BatchNorm2d(64)
elif self.modalbn == 3:
self.bn1_ir = nn.BatchNorm2d(64)
self.bn1_modalx = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
# else:
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=last_conv_stride, dilation=last_conv_dilation)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def _make_layer(self, block, planes, blocks, stride=1, dilation=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample, dilation, isshape=self.isshape, modalbn=self.modalbn))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes, isshape=self.isshape, modalbn=self.modalbn))
return Sequential(*layers)
def forward(self, x, modal=0):
x = self.conv1(x)
if modal == 0: # RGB
bbn1 = self.bn1
elif modal == 1: # IR
bbn1 = self.bn1_ir
elif modal == 2: # modalx
bbn1 = self.bn1_modalx
elif modal == 3: # shape
assert self.isshape == True
bbn1 = self.bn1_shape
x = bbn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x, modal)
x = self.layer2(x, modal)
x = self.layer3(x, modal)
x = self.layer4(x, modal)
return x
def init_bn_layer(self, layer):
for i in range(len(layer)):
if self.isshape:
layer[i].bn1_shape.weight.data = layer[i].bn1.weight.data.clone()
layer[i].bn1_shape.bias.data = layer[i].bn1.bias.data.clone()
layer[i].bn2_shape.weight.data = layer[i].bn2.weight.data.clone()
layer[i].bn2_shape.bias.data = layer[i].bn2.bias.data.clone()
layer[i].bn3_shape.weight.data = layer[i].bn3.weight.data.clone()
layer[i].bn3_shape.bias.data = layer[i].bn3.bias.data.clone()
if layer[i].downsample is not None:
layer[i].dsbn_shape.weight.data = layer[i].downsample[1].weight.data.clone()
layer[i].dsbn_shape.bias.data = layer[i].downsample[1].bias.data.clone()
if self.modalbn >= 2:
layer[i].bn1_ir.weight.data = layer[i].bn1.weight.data.clone()
layer[i].bn1_ir.bias.data = layer[i].bn1.bias.data.clone()
layer[i].bn2_ir.weight.data = layer[i].bn2.weight.data.clone()
layer[i].bn2_ir.bias.data = layer[i].bn2.bias.data.clone()
layer[i].bn3_ir.weight.data = layer[i].bn3.weight.data.clone()
layer[i].bn3_ir.bias.data = layer[i].bn3.bias.data.clone()
if layer[i].downsample is not None:
layer[i].dsbn_ir.weight.data = layer[i].downsample[1].weight.data.clone()
layer[i].dsbn_ir.bias.data = layer[i].downsample[1].bias.data.clone()
if self.modalbn == 3:
layer[i].bn1_modalx.weight.data = layer[i].bn1.weight.data.clone()
layer[i].bn1_modalx.bias.data = layer[i].bn1.bias.data.clone()
layer[i].bn2_modalx.weight.data = layer[i].bn2.weight.data.clone()
layer[i].bn2_modalx.bias.data = layer[i].bn2.bias.data.clone()
layer[i].bn3_modalx.weight.data = layer[i].bn3.weight.data.clone()
layer[i].bn3_modalx.bias.data = layer[i].bn3.bias.data.clone()
if layer[i].downsample is not None:
layer[i].dsbn_modalx.weight.data = layer[i].downsample[1].weight.data.clone()
layer[i].dsbn_modalx.bias.data = layer[i].downsample[1].bias.data.clone()
def init_bn(self, onlyshallow=False):
# if onlyshallow:
if self.isshape:
self.bn1_shape.weight.data = self.bn1.weight.data.clone()
self.bn1_shape.bias.data = self.bn1.bias.data.clone()
if self.modalbn >= 2:
self.bn1_ir.weight.data = self.bn1.weight.data.clone()
self.bn1_ir.bias.data = self.bn1.bias.data.clone()
if self.modalbn == 3:
self.bn1_modalx.weight.data = self.bn1.weight.data.clone()
self.bn1_modalx.bias.data = self.bn1.bias.data.clone()
# else:
self.init_bn_layer(self.layer1)
self.init_bn_layer(self.layer2)
self.init_bn_layer(self.layer3)
self.init_bn_layer(self.layer4)
def average_bn_layer(self, layer):
for i in range(len(layer)):
bn1w = (layer[i].bn1.weight.data.clone()+layer[i].bn1_ir.weight.data.clone()+layer[i].bn1_shape.weight.data.clone())/3
bn1b = (layer[i].bn1.bias.data.clone()+layer[i].bn1_ir.bias.data.clone()+layer[i].bn1_shape.bias.data.clone())/3
bn2w = (layer[i].bn2.weight.data.clone()+layer[i].bn2_ir.weight.data.clone()+layer[i].bn2_shape.weight.data.clone())/3
bn2b = (layer[i].bn2.bias.data.clone()+layer[i].bn2_ir.bias.data.clone()+layer[i].bn2_shape.bias.data.clone())/3
bn3w = (layer[i].bn3.weight.data.clone()+layer[i].bn3_ir.weight.data.clone()+layer[i].bn3_shape.weight.data.clone())/3
bn3b = (layer[i].bn3.bias.data.clone()+layer[i].bn3_ir.bias.data.clone()+layer[i].bn3_shape.bias.data.clone())/3
if layer[i].downsample is not None:
dbbnw = (layer[i].downsample[1].weight.data.clone()+layer[i].dsbn_shape.weight.data.clone()+layer[i].dsbn_ir.weight.data.clone())/3
dbbnb = (layer[i].downsample[1].bias.data.clone()+layer[i].dsbn_shape.bias.data.clone()+layer[i].dsbn_ir.bias.data.clone())/3
layer[i].bn1.weight.data = bn1w.clone()
layer[i].bn1.bias.data = bn1b.clone()
layer[i].bn2.weight.data = bn2w.clone()
layer[i].bn2.bias.data = bn2b.clone()
layer[i].bn3.weight.data = bn3w.clone()
layer[i].bn3.bias.data = bn3b.clone()
if layer[i].downsample is not None:
layer[i].downsample[1].weight.data = dbbnw.clone()
layer[i].downsample[1].bias.data = dbbnb.clone()
if self.isshape:
layer[i].bn1_shape.weight.data = bn1w.clone()
layer[i].bn1_shape.bias.data = bn1b.clone()
layer[i].bn2_shape.weight.data = bn2w.clone()
layer[i].bn2_shape.bias.data = bn2b.clone()
layer[i].bn3_shape.weight.data = bn3w.clone()
layer[i].bn3_shape.bias.data = bn3b.clone()
if layer[i].downsample is not None:
layer[i].dsbn_shape.weight.data = dbbnw.clone()
layer[i].dsbn_shape.bias.data = dbbnb.clone()
if self.modalbn >= 2:
layer[i].bn1_ir.weight.data = bn1w.clone()
layer[i].bn1_ir.bias.data = bn1b.clone()
layer[i].bn2_ir.weight.data = bn2w.clone()
layer[i].bn2_ir.bias.data = bn2b.clone()
layer[i].bn3_ir.weight.data = bn3w.clone()
layer[i].bn3_ir.bias.data = bn3b.clone()
if layer[i].downsample is not None:
layer[i].dsbn_ir.weight.data = dbbnw.clone()
layer[i].dsbn_ir.bias.data = dbbnb.clone()
if self.modalbn == 3:
layer[i].bn1_modalx.weight.data = bn1w.clone()
layer[i].bn1_modalx.bias.data = bn1b.clone()
layer[i].bn2_modalx.weight.data = bn2w.clone()
layer[i].bn2_modalx.bias.data = bn2b.clone()
layer[i].bn3_modalx.weight.data = bn3w.clone()
layer[i].bn3_modalx.bias.data = bn3b.clone()
if layer[i].downsample is not None:
layer[i].dsbn_modalx.weight.data = dbbnw.clone()
layer[i].dsbn_modalx.bias.data = dbbnb.clone()
def average_bn(self, onlyshallow=False):
if onlyshallow:
tmpw = self.bn1.weight.data.clone() + self.bn1_shape.weight.data.clone()
tmpw /= 2
tmpb = self.bn1.bias.data.clone() + self.bn1_shape.bias.data.clone()
tmpb /= 2
self.bn1.weight.data = tmpw.clone()
self.bn1.bias.data = tmpb.clone()
if self.isshape:
self.bn1_shape.weight.data = tmpw.clone()
self.bn1_shape.bias.data = tmpb.clone()
if self.modalbn >= 2:
self.bn1_ir.weight.data = tmpw.clone()
self.bn1_ir.bias.data = tmpb.clone()
if self.modalbn == 3:
self.bn1_modalx.weight.data = tmpw.clone()
self.bn1_modalx.bias.data = tmpb.clone()
else:
self.average_bn_layer(self.layer1)
self.average_bn_layer(self.layer2)
self.average_bn_layer(self.layer3)
self.average_bn_layer(self.layer4)
# def last_layer_shared(self, ):
# res = []
# for k, v in self.layer4.named_parameters():
# if 'conv' in k:
# res.append(v)
# return res
def remove_fc(state_dict):
"""Remove the fc layer parameters from state_dict."""
# for key, value in state_dict.items():
for key, value in list(state_dict.items()):
if key.startswith('fc.'):
del state_dict[key]
return state_dict
def resnet18(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
if pretrained:
model.load_state_dict(remove_fc(model_zoo.load_url(model_urls['resnet18'])))
return model
def resnet34(pretrained=False, **kwargs):
"""Constructs a ResNet-34 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
if pretrained:
model.load_state_dict(remove_fc(model_zoo.load_url(model_urls['resnet34'])))
return model
def resnet50(pretrained=False, **kwargs):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
if pretrained:
# model.load_state_dict(remove_fc(model_zoo.load_url(model_urls['resnet50'])))
model.load_state_dict(remove_fc(model_zoo.load_url(model_urls['resnet50'])),strict=False)
return model
def resnet101(pretrained=False, **kwargs):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
if pretrained:
model.load_state_dict(
remove_fc(model_zoo.load_url(model_urls['resnet101'])),strict=False)
return model
def resnet152(pretrained=False, **kwargs):
"""Constructs a ResNet-152 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
if pretrained:
model.load_state_dict(
remove_fc(model_zoo.load_url(model_urls['resnet152'])), strict=False)
return model