File size: 12,960 Bytes
d526dbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Original Author: Jonathan Donnellya (jonathan.donnelly@maine.edu)
# Modified by Mohammad Reza Taesiri (mtaesiri@gmail.com)

import os
import torch
import torch.nn as nn
from collections import OrderedDict

model_dir = os.path.dirname(os.path.realpath(__file__))


def conv3x3(in_planes, out_planes, stride=1):
    """3x3 convolution with padding"""
    return nn.Conv2d(
        in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False
    )


def conv1x1(in_planes, out_planes, stride=1):
    """1x1 convolution"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)


class BasicBlock(nn.Module):
    # class attribute
    expansion = 1
    num_layers = 2

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        # only conv with possibly not 1 stride
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes)

        # if stride is not 1 then self.downsample cannot be None
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = 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:
            identity = self.downsample(x)

        # the residual connection
        out += identity
        out = self.relu(out)

        return out

    def block_conv_info(self):
        block_kernel_sizes = [3, 3]
        block_strides = [self.stride, 1]
        block_paddings = [1, 1]

        return block_kernel_sizes, block_strides, block_paddings


class Bottleneck(nn.Module):
    # class attribute
    expansion = 4
    num_layers = 3

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = conv1x1(inplanes, planes)
        self.bn1 = nn.BatchNorm2d(planes)
        # only conv with possibly not 1 stride
        self.conv2 = conv3x3(planes, planes, stride)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = conv1x1(planes, planes * self.expansion)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)

        # if stride is not 1 then self.downsample cannot be None
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu(out)

        return out

    def block_conv_info(self):
        block_kernel_sizes = [1, 3, 1]
        block_strides = [1, self.stride, 1]
        block_paddings = [0, 1, 0]

        return block_kernel_sizes, block_strides, block_paddings


class ResNet_features(nn.Module):
    """
    the convolutional layers of ResNet
    the average pooling and final fully convolutional layer is removed
    """

    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False):
        super(ResNet_features, self).__init__()

        self.inplanes = 64

        # the first convolutional layer before the structured sequence of blocks
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        # comes from the first conv and the following max pool
        self.kernel_sizes = [7, 3]
        self.strides = [2, 2]
        self.paddings = [3, 1]

        # the following layers, each layer is a sequence of blocks
        self.block = block
        self.layers = layers
        self.layer1 = self._make_layer(
            block=block, planes=64, num_blocks=self.layers[0]
        )
        self.layer2 = self._make_layer(
            block=block, planes=128, num_blocks=self.layers[1], stride=2
        )
        self.layer3 = self._make_layer(
            block=block, planes=256, num_blocks=self.layers[2], stride=2
        )
        self.layer4 = self._make_layer(
            block=block, planes=512, num_blocks=self.layers[3], stride=2
        )

        # initialize the parameters
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        # Zero-initialize the last BN in each residual branch,
        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def _make_layer(self, block, planes, num_blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        # only the first block has downsample that is possibly not None
        layers.append(block(self.inplanes, planes, stride, downsample))

        self.inplanes = planes * block.expansion
        for _ in range(1, num_blocks):
            layers.append(block(self.inplanes, planes))

        # keep track of every block's conv size, stride size, and padding size
        for each_block in layers:
            (
                block_kernel_sizes,
                block_strides,
                block_paddings,
            ) = each_block.block_conv_info()
            self.kernel_sizes.extend(block_kernel_sizes)
            self.strides.extend(block_strides)
            self.paddings.extend(block_paddings)

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        return x

    def conv_info(self):
        return self.kernel_sizes, self.strides, self.paddings

    def num_layers(self):
        """
        the number of conv layers in the network, not counting the number
        of bypass layers
        """

        return (
            self.block.num_layers * self.layers[0]
            + self.block.num_layers * self.layers[1]
            + self.block.num_layers * self.layers[2]
            + self.block.num_layers * self.layers[3]
            + 1
        )

    def __repr__(self):
        template = "resnet{}_features"
        return template.format(self.num_layers() + 1)


def resnet50_features(pretrained=True, inat=True, **kwargs):
    """Constructs a ResNet-50 model.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet or iNaturalist
        pretrained (bool): If True, returns a model pre-trained on iNaturalst; else, ImageNet
    """
    model = ResNet_features(Bottleneck, [3, 4, 6, 4], **kwargs)
    if pretrained:
        if inat:
            # print('Loading iNat model')
            model_dict = torch.load(
                model_dir
                + "/../../weights/"
                + "BBN.iNaturalist2017.res50.90epoch.best_model.pth.pt"
            )
        else:
            raise

        if inat:
            model_dict.pop("module.classifier.weight")
            model_dict.pop("module.classifier.bias")
            for key in list(model_dict.keys()):
                model_dict[
                    key.replace("module.backbone.", "")
                    .replace("cb_block", "layer4.2")
                    .replace("rb_block", "layer4.3")
                ] = model_dict.pop(key)

        else:
            raise

        model.load_state_dict(model_dict, strict=False)

    return model


class ResNet_classifier(nn.Module):
    """
    A classifier for Deformable ProtoPNet
    """

    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False):
        super(ResNet_classifier, self).__init__()

        self.inplanes = 64

        # the first convolutional layer before the structured sequence of blocks
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        # comes from the first conv and the following max pool
        self.kernel_sizes = [7, 3]
        self.strides = [2, 2]
        self.paddings = [3, 1]

        # the following layers, each layer is a sequence of blocks
        self.block = block
        self.layers = layers
        self.layer1 = self._make_layer(
            block=block, planes=64, num_blocks=self.layers[0]
        )
        self.layer2 = self._make_layer(
            block=block, planes=128, num_blocks=self.layers[1], stride=2
        )
        self.layer3 = self._make_layer(
            block=block, planes=256, num_blocks=self.layers[2], stride=2
        )
        self.layer4 = self._make_layer(
            block=block, planes=512, num_blocks=self.layers[3], stride=2
        )

        self.classifier = nn.Linear(2048 * 7 * 7, 200)

        # initialize the parameters
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        # Zero-initialize the last BN in each residual branch,
        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def _make_layer(self, block, planes, num_blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        # only the first block has downsample that is possibly not None
        layers.append(block(self.inplanes, planes, stride, downsample))

        self.inplanes = planes * block.expansion
        for _ in range(1, num_blocks):
            layers.append(block(self.inplanes, planes))

        # keep track of every block's conv size, stride size, and padding size
        for each_block in layers:
            (
                block_kernel_sizes,
                block_strides,
                block_paddings,
            ) = each_block.block_conv_info()
            self.kernel_sizes.extend(block_kernel_sizes)
            self.strides.extend(block_strides)
            self.paddings.extend(block_paddings)

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.classifier(torch.flatten(x, start_dim=1))
        return x

    def conv_info(self):
        return self.kernel_sizes, self.strides, self.paddings

    def num_layers(self):
        """
        the number of conv layers in the network, not counting the number
        of bypass layers
        """

        return (
            self.block.num_layers * self.layers[0]
            + self.block.num_layers * self.layers[1]
            + self.block.num_layers * self.layers[2]
            + self.block.num_layers * self.layers[3]
            + 1
        )

    def __repr__(self):
        template = "resnet{}_features"
        return template.format(self.num_layers() + 1)