-
Notifications
You must be signed in to change notification settings - Fork 1
/
multi_modal_model.py
184 lines (153 loc) · 6.08 KB
/
multi_modal_model.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.init as init
class View(nn.Module):
def __init__(self, size):
super(View, self).__init__()
self.size = size
def forward(self, tensor):
return tensor.view(self.size)
class VAE(nn.Module):
"""Encoder-Decoder architecture for both WAE-MMD and WAE-GAN."""
def __init__(self, z_dim=32, nc=1):
super(VAE, self).__init__()
self.z_dim = z_dim
self.nc = nc
self.encoder = nn.Sequential(
nn.Conv2d(nc, 64, 4, 2, 1, bias=False), # B, 64, 64, 64
nn.BatchNorm2d(64),
nn.ReLU(True),
nn.Conv2d(64, 128, 4, 2, 1, bias=False), # B, 128, 32, 32
nn.BatchNorm2d(128),
nn.ReLU(True),
nn.Conv2d(128, 256, 4, 2, 1, bias=False), # B, 256, 16, 16
nn.BatchNorm2d(256),
nn.ReLU(True),
nn.Conv2d(256, 512, 4, 2, 1, bias=False), # B, 512, 8, 8
nn.BatchNorm2d(512),
nn.ReLU(True),
nn.Conv2d(512, 1024, 4, 2, 1, bias=False), # B, 1024, 4, 4
nn.BatchNorm2d(1024),
nn.ReLU(True),
View((-1, 1024*4*4)), # B, 1024*4*4
)
self.fc_mu = nn.Linear(1024*4*4, z_dim) # B, z_dim
self.fc_logvar = nn.Linear(1024*4*4, z_dim) # B, z_dim
self.decoder = nn.Sequential(
nn.Linear(z_dim, 1024*4*4), # B, 1024*4*4
View((-1, 1024, 4, 4)), # B, 1024, 4, 4
nn.ConvTranspose2d(1024, 512, 4, 2, 1, bias=False), # B, 512, 8, 8
nn.BatchNorm2d(512),
nn.ReLU(True),
nn.ConvTranspose2d(512, 256, 4, 2, 1, bias=False), # B, 256, 16, 16
nn.BatchNorm2d(256),
nn.ReLU(True),
nn.ConvTranspose2d(256, 128, 4, 2, 1, bias=False), # B, 128, 32, 32
nn.BatchNorm2d(128),
nn.ReLU(True),
nn.ConvTranspose2d(128, 64, 4, 2, 1, bias=False), # B, 64, 64, 64
nn.BatchNorm2d(64),
nn.ReLU(True),
nn.ConvTranspose2d(64, 32, 4, 2, 1, bias=False), # B, 32, 128, 128
nn.BatchNorm2d(32),
nn.ReLU(True),
nn.ConvTranspose2d(32, nc, 1) # B, nc, 256, 256
)
self.aux_decoder = nn.Sequential(
nn.Linear(z_dim, 1024*4*4), # B, 1024*4*4
View((-1, 1024, 4, 4)), # B, 1024, 4, 4
nn.ConvTranspose2d(1024, 512, 4, 2, 1, bias=False), # B, 512, 8, 8
nn.BatchNorm2d(512),
nn.ReLU(True),
nn.ConvTranspose2d(512, 256, 4, 2, 1, bias=False), # B, 256, 16, 16
nn.BatchNorm2d(256),
nn.ReLU(True),
nn.ConvTranspose2d(256, 128, 4, 2, 1, bias=False), # B, 128, 32, 32
nn.BatchNorm2d(128),
nn.ReLU(True),
nn.ConvTranspose2d(128, 64, 4, 2, 1, bias=False), # B, 64, 64, 64
nn.BatchNorm2d(64),
nn.ReLU(True),
nn.ConvTranspose2d(64, 32, 4, 2, 1, bias=False), # B, 32, 128, 128
nn.BatchNorm2d(32),
nn.ReLU(True),
nn.ConvTranspose2d(32, nc, 1) # B, 1, 256, 256
)
self.weight_init()
def weight_init(self):
for block in self._modules:
try:
for m in self._modules[block]:
kaiming_init(m)
except:
kaiming_init(block)
def forward(self, x):
z = self._encode(x)
mu, logvar = self.fc_mu(z), self.fc_logvar(z)
z = self.reparameterize(mu, logvar)
x_recon = self._decode(z)
x_aux = self._aux_decode(z)
return x_recon, x_aux, z, mu, logvar
def reparameterize(self, mu, logvar):
stds = (0.5 * logvar).exp()
epsilon = torch.randn(*mu.size())
if mu.is_cuda:
stds, epsilon = stds.cuda(), epsilon.cuda()
latents = epsilon * stds + mu
return latents
def _encode(self, x):
return self.encoder(x)
def _decode(self, z):
return self.decoder(z)
def _aux_decode(self,z):
return self.aux_decoder(z)
class Discriminator(nn.Module):
"""Adversary architecture(Discriminator) for WAE-GAN."""
def __init__(self, z_dim=10):
super(Discriminator, self).__init__()
self.z_dim = z_dim
self.net = nn.Sequential(
nn.Linear(z_dim, 128),
nn.BatchNorm1d(128),
nn.ReLU(True),
nn.Linear(128,256),
nn.BatchNorm1d(256),
nn.ReLU(True),
nn.Linear(256,512),
nn.BatchNorm1d(512),
nn.ReLU(True),
nn.Linear(512, 256),
nn.BatchNorm1d(256),
nn.ReLU(True),
nn.Linear(256, 128),
nn.BatchNorm1d(128),
nn.ReLU(True),
nn.Linear(128, 1),
nn.Sigmoid()
)
self.weight_init()
def weight_init(self):
for block in self._modules:
for m in self._modules[block]:
kaiming_init(m)
def forward(self, z):
return self.net(z)
def kaiming_init(m):
if isinstance(m, (nn.Linear, nn.Conv2d)):
init.kaiming_normal(m.weight)
if m.bias is not None:
m.bias.data.fill_(0)
elif isinstance(m, (nn.BatchNorm1d, nn.BatchNorm2d)):
m.weight.data.fill_(1)
if m.bias is not None:
m.bias.data.fill_(0)
def normal_init(m, mean, std):
if isinstance(m, (nn.Linear, nn.Conv2d)):
m.weight.data.normal_(mean, std)
if m.bias.data is not None:
m.bias.data.zero_()
elif isinstance(m, (nn.BatchNorm2d, nn.BatchNorm1d)):
m.weight.data.fill_(1)
if m.bias.data is not None:
m.bias.data.zero_()