File size: 8,880 Bytes
8d0209c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
import torch.nn.functional as F
import math


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


class ConvBlock(nn.Module):
    def __init__(self, in_planes, out_planes):
        super(ConvBlock, self).__init__()
        self.bn1 = nn.BatchNorm2d(in_planes)
        self.conv1 = conv3x3(in_planes, int(out_planes / 2))
        self.bn2 = nn.BatchNorm2d(int(out_planes / 2))
        self.conv2 = conv3x3(int(out_planes / 2), int(out_planes / 4))
        self.bn3 = nn.BatchNorm2d(int(out_planes / 4))
        self.conv3 = conv3x3(int(out_planes / 4), int(out_planes / 4))

        if in_planes != out_planes:
            self.downsample = nn.Sequential(
                nn.BatchNorm2d(in_planes),
                nn.ReLU(True),
                nn.Conv2d(in_planes, out_planes,
                          kernel_size=1, stride=1, bias=False),
            )
        else:
            self.downsample = None

    def forward(self, x):
        residual = x

        out1 = self.bn1(x)
        out1 = F.relu(out1, True)
        out1 = self.conv1(out1)

        out2 = self.bn2(out1)
        out2 = F.relu(out2, True)
        out2 = self.conv2(out2)

        out3 = self.bn3(out2)
        out3 = F.relu(out3, True)
        out3 = self.conv3(out3)

        out3 = torch.cat((out1, out2, out3), 1)

        if self.downsample is not None:
            residual = self.downsample(residual)

        out3 += residual

        return out3


class Bottleneck(nn.Module):

    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, 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, planes * 4, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * 4)
        self.relu = nn.ReLU(inplace=True)
        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)
        out = self.relu(out)

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

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

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

        return out


class HourGlass(nn.Module):
    def __init__(self, num_modules, depth, num_features):
        super(HourGlass, self).__init__()
        self.num_modules = num_modules
        self.depth = depth
        self.features = num_features

        self._generate_network(self.depth)

    def _generate_network(self, level):
        self.add_module('b1_' + str(level), ConvBlock(self.features, self.features))

        self.add_module('b2_' + str(level), ConvBlock(self.features, self.features))

        if level > 1:
            self._generate_network(level - 1)
        else:
            self.add_module('b2_plus_' + str(level), ConvBlock(self.features, self.features))

        self.add_module('b3_' + str(level), ConvBlock(self.features, self.features))

    def _forward(self, level, inp):
        # Upper branch
        up1 = inp
        up1 = self._modules['b1_' + str(level)](up1)

        # Lower branch
        low1 = F.avg_pool2d(inp, 2, stride=2)
        low1 = self._modules['b2_' + str(level)](low1)

        if level > 1:
            low2 = self._forward(level - 1, low1)
        else:
            low2 = low1
            low2 = self._modules['b2_plus_' + str(level)](low2)

        low3 = low2
        low3 = self._modules['b3_' + str(level)](low3)

        up2 = F.interpolate(low3, scale_factor=2, mode='nearest')

        return up1 + up2

    def forward(self, x):
        return self._forward(self.depth, x)


class FAN(nn.Module):

    def __init__(self, num_modules=1):
        super(FAN, self).__init__()
        self.num_modules = num_modules

        # Base part
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3)
        self.bn1 = nn.BatchNorm2d(64)
        self.conv2 = ConvBlock(64, 128)
        self.conv3 = ConvBlock(128, 128)
        self.conv4 = ConvBlock(128, 256)

        # Stacking part
        for hg_module in range(self.num_modules):
            self.add_module('m' + str(hg_module), HourGlass(1, 4, 256))
            self.add_module('top_m_' + str(hg_module), ConvBlock(256, 256))
            self.add_module('conv_last' + str(hg_module),
                            nn.Conv2d(256, 256, kernel_size=1, stride=1, padding=0))
            self.add_module('bn_end' + str(hg_module), nn.BatchNorm2d(256))
            self.add_module('l' + str(hg_module), nn.Conv2d(256,
                                                            68, kernel_size=1, stride=1, padding=0))

            if hg_module < self.num_modules - 1:
                self.add_module(
                    'bl' + str(hg_module), nn.Conv2d(256, 256, kernel_size=1, stride=1, padding=0))
                self.add_module('al' + str(hg_module), nn.Conv2d(68,
                                                                 256, kernel_size=1, stride=1, padding=0))

    def forward(self, x):
        x = F.relu(self.bn1(self.conv1(x)), True)
        x = F.avg_pool2d(self.conv2(x), 2, stride=2)
        x = self.conv3(x)
        x = self.conv4(x)

        previous = x

        outputs = []
        for i in range(self.num_modules):
            hg = self._modules['m' + str(i)](previous)

            ll = hg
            ll = self._modules['top_m_' + str(i)](ll)

            ll = F.relu(self._modules['bn_end' + str(i)]
                        (self._modules['conv_last' + str(i)](ll)), True)

            # Predict heatmaps
            tmp_out = self._modules['l' + str(i)](ll)
            outputs.append(tmp_out)

            if i < self.num_modules - 1:
                ll = self._modules['bl' + str(i)](ll)
                tmp_out_ = self._modules['al' + str(i)](tmp_out)
                previous = previous + ll + tmp_out_

        return outputs


class ResNetDepth(nn.Module):

    def __init__(self, block=Bottleneck, layers=[3, 8, 36, 3], num_classes=68):
        self.inplanes = 64
        super(ResNetDepth, self).__init__()
        self.conv1 = nn.Conv2d(3 + 68, 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)
        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=2)
        self.avgpool = nn.AvgPool2d(7)
        self.fc = nn.Linear(512 * block.expansion, num_classes)

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

        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.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.fc(x)

        return x