chairs/models_cvae.py at master · jnnan/chairs · GitHub

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
import torch
import torch.nn as nn
import torch.nn.functional as F


class CVAE(nn.Module):
    def __init__(self, in_channels=1, dim=64, out_conv_channels=512, hidden_dim=64):
        super(CVAE, self).__init__()
        self.encoder = Encoder(in_channels, dim, out_conv_channels, hidden_dim)

        self.decoder = Decoder(out_conv_channels, dim, in_channels, hidden_dim )

    def reparameterize(self, mu, log_var):
        std = torch.exp(log_var / 2)
        if self.training:
            eps = torch.randn_like(std)
            return mu + eps * std
        else:
            return mu

    def forward(self, x, x_human):
        mu, log_var = self.encoder(x, x_human)
        z = self.reparameterize(mu, log_var)
        x_rec = self.decoder(z, x_human)
        return x_rec, mu, log_var, z


class Encoder(torch.nn.Module):
    def __init__(self, in_channels=1, dim=64, out_conv_channels=512, hidden_dim=64):
        super(Encoder, self).__init__()
        conv1_channels = int(out_conv_channels / 8)
        conv2_channels = int(out_conv_channels / 4)
        conv3_channels = int(out_conv_channels / 2)
        self.out_conv_channels = out_conv_channels
        self.dim = dim
        self.out_dim = int(dim / 16)
        self.hidden_dim = hidden_dim

        self.conv1 = nn.Sequential(
            nn.Conv3d(
                in_channels=in_channels + 1, out_channels=conv1_channels, kernel_size=4,
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv1_channels),
            nn.LeakyReLU(0.2, inplace=True)
        )
        self.conv2 = nn.Sequential(
            nn.Conv3d(
                in_channels=conv1_channels + 1, out_channels=conv2_channels, kernel_size=4,
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv2_channels),
            nn.LeakyReLU(0.2, inplace=True)
        )
        self.conv3 = nn.Sequential(
            nn.Conv3d(
                in_channels=conv2_channels + 1, out_channels=conv3_channels, kernel_size=4,
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv3_channels),
            nn.LeakyReLU(0.2, inplace=True)
        )
        self.conv4 = nn.Sequential(
            nn.Conv3d(
                in_channels=conv3_channels + 1, out_channels=out_conv_channels, kernel_size=4,
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(out_conv_channels),
            nn.LeakyReLU(0.2, inplace=True)
        )
        self.mean_out = nn.Sequential(
            nn.Linear(out_conv_channels * self.out_dim * self.out_dim * self.out_dim, self.hidden_dim),
            #nn.Sigmoid(),
        )
        self.var_out = nn.Sequential(
            nn.Linear(out_conv_channels * self.out_dim * self.out_dim * self.out_dim, self.hidden_dim),
            #nn.Sigmoid(),
        )

    def forward(self, x, x_human):
        x = torch.cat([x, x_human], dim=1)
        x = self.conv1(x)
        x = torch.cat([x, F.interpolate(x_human, (self.dim // 2, self.dim // 2, self.dim // 2))], dim=1)
        x = self.conv2(x)
        x = torch.cat([x, F.interpolate(x_human, (self.dim // 4, self.dim // 4, self.dim // 4))], dim=1)
        x = self.conv3(x)
        x = torch.cat([x, F.interpolate(x_human, (self.dim // 8, self.dim // 8, self.dim // 8))], dim=1)
        x = self.conv4(x)
        # Flatten and apply linear + sigmoid
        x = x.view(-1, self.out_conv_channels * self.out_dim * self.out_dim * self.out_dim)
        mean = self.mean_out(x)
        log_var = self.var_out(x)
        return mean, log_var


class Decoder(torch.nn.Module):
    def __init__(self, in_channels=512, out_dim=64, out_channels=1, noise_dim=64, activation="sigmoid"):
        super(Decoder, self).__init__()
        self.in_channels = in_channels
        self.out_dim = out_dim
        self.in_dim = int(out_dim / 16)
        conv1_out_channels = int(self.in_channels / 2.0)
        conv2_out_channels = int(conv1_out_channels / 2)
        conv3_out_channels = int(conv2_out_channels / 2)
        conv4_out_channels = int(conv3_out_channels / 2)

        self.linear = torch.nn.Linear(noise_dim, in_channels * self.in_dim * self.in_dim * self.in_dim)

        self.conv1 = nn.Sequential(
            nn.ConvTranspose3d(
                in_channels=in_channels, out_channels=conv1_out_channels, kernel_size=(4, 4, 4),
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv1_out_channels),
            nn.ReLU(inplace=True)
        )
        self.conv2 = nn.Sequential(
            nn.ConvTranspose3d(
                in_channels=conv1_out_channels + 1, out_channels=conv2_out_channels, kernel_size=(4, 4, 4),
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv2_out_channels),
            nn.ReLU(inplace=True)
        )
        self.conv3 = nn.Sequential(
            nn.ConvTranspose3d(
                in_channels=conv2_out_channels + 1, out_channels=conv3_out_channels, kernel_size=(4, 4, 4),
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv3_out_channels),
            nn.ReLU(inplace=True)
        )
        self.conv4 = nn.Sequential(
            nn.ConvTranspose3d(
                in_channels=conv3_out_channels + 1, out_channels=conv4_out_channels, kernel_size=(4, 4, 4),
                stride=2, padding=1, bias=False
            ),
            nn.BatchNorm3d(conv4_out_channels),
            nn.ReLU(inplace=True)
        )
        self.conv_out_1 = nn.Sequential(
            nn.Conv3d(
                in_channels=conv4_out_channels + 1, out_channels=conv4_out_channels, kernel_size=(1, 1, 1),
                stride=1, padding=0, bias=False
            ),
            nn.BatchNorm3d(conv4_out_channels),
            nn.ReLU(inplace=True)
        )
        self.conv_out_2= nn.Sequential(
            nn.Conv3d(
                in_channels=conv4_out_channels, out_channels=out_channels, kernel_size=(1, 1, 1),
                stride=1, padding=0, bias=False
            ),
        )
        if activation == "sigmoid":
            self.out = torch.nn.Sigmoid()
        else:
            self.out = torch.nn.Tanh()

    def project(self, x):
        """
        projects and reshapes latent vector to starting volume
        :param x: latent vector
        :return: starting volume
        """
        return x.view(-1, self.in_channels, self.in_dim, self.in_dim, self.in_dim)

    def forward(self, x, x_human):
        x = self.linear(x)
        x = self.project(x)
        # x = torch.cat([x, F.interpolate(x_human, (self.out_dim // 16, self.out_dim // 16, self.out_dim // 16))], dim=1)
        x = self.conv1(x)
        x = torch.cat([x, F.interpolate(x_human, (self.out_dim // 8, self.out_dim // 8, self.out_dim // 8))], dim=1)
        x = self.conv2(x)
        x = torch.cat([x, F.interpolate(x_human, (self.out_dim // 4, self.out_dim // 4, self.out_dim // 4))], dim=1)
        x = self.conv3(x)
        x = torch.cat([x, F.interpolate(x_human, (self.out_dim // 2, self.out_dim // 2, self.out_dim // 2))], dim=1)
        x = self.conv4(x)
        x = torch.cat([x, F.interpolate(x_human, (self.out_dim // 1, self.out_dim // 1, self.out_dim // 1))], dim=1)
        x = self.conv_out_1(x)
        x = self.conv_out_2(x)
        return self.out(x)