-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel.py
199 lines (158 loc) · 6.41 KB
/
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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
import torch
import torch.nn as nn
##### Transpose is learning parameter while Up-sampling is no-learning parameters.
##### Using Up-samling for faster inference or
##### training because it does not require to update weight or compute gradient
#########################
## CoGAN
#########################
#########################
## Discriminator Net
#########################
class CoDisMNIST(nn.Module): ## shared_weights
def __init__(self):
super(CoDisMNIST, self).__init__()
## Domain A ##
self.conv0_a = nn.Conv2d(in_channels=1, out_channels=20, kernel_size=5, stride=1, padding=0)
self.pool0_a = nn.MaxPool2d(kernel_size=2)
## Domain B ##
self.conv0_b = nn.Conv2d(in_channels=1, out_channels=20, kernel_size=5, stride=1, padding=0)
self.pool0_b = nn.MaxPool2d(kernel_size=2)
self.shared_weight = nn.Sequential(
nn.Conv2d(in_channels=20, out_channels=50, kernel_size=5, stride=1, padding=0),
nn.MaxPool2d(kernel_size=2)
)
## Shared convolution ##
self.D = nn.Sequential(
nn.Linear(3200, 500),
nn.PReLU(),
nn.Linear(500,1)
)
self.sigmoid = nn.Sigmoid()
def forward(self, xa, xb):
h0_a = self.pool0_a(self.conv0_a(xa))
h0_b = self.pool0_b(self.conv0_b(xb))
hc = torch.cat((h0_a, h0_b), 0)
h = self.shared_weight(hc)
out = h.view(h.shape[0], -1)
out = self.sigmoid(self.D(out))
h_a = out[:xa.shape[0]]
h_b = out[xa.shape[0]:]
return h_a, h_b
###########################
####### Generator Net #####
###########################
class CoGenMNIST(nn.Module):
def __init__(self, latend_dims):
super(CoGenMNIST, self).__init__()
self.latent_dims = latend_dims
self.shared_conv = nn.Sequential(
nn.ConvTranspose2d(in_channels=self.latent_dims, out_channels=1024, kernel_size=4, stride = 1),
nn.BatchNorm2d(1024),
nn.PReLU(),
nn.ConvTranspose2d(in_channels=1024, out_channels=512, kernel_size=3, stride = 2),
nn.BatchNorm2d(512),
nn.PReLU(),
nn.ConvTranspose2d(in_channels=512, out_channels=256, kernel_size=3, stride = 2),
nn.BatchNorm2d(256),
nn.PReLU(),
nn.ConvTranspose2d(in_channels=256, out_channels=128, kernel_size=3, stride = 2),
nn.BatchNorm2d(128),
nn.PReLU(),
)
self.G1 = nn.Sequential(
nn.ConvTranspose2d(in_channels=128, out_channels=1, kernel_size=6, stride = 1),
nn.PReLU()
)
self.G2 = nn.Sequential(
nn.ConvTranspose2d(in_channels=128, out_channels=1, kernel_size=6, stride = 1),
nn.PReLU()
)
self.tanh = nn.Tanh()
def forward(self, z):
z = z.view(z.size(0), z.size(1), 1, 1) # noise vector : z
h = self.shared_conv(z)
out_a = self.tanh(self.G1(h))
out_b = self.tanh(self.G2(h))
return out_a, out_b
##################################################
## CoGAN - rotation
##################################################
class R_CoGANGenMNIST(nn.Module):
def __init__(self):
super(R_CoGANGenMNIST, self).__init__()
self.sw = nn.Sequential(
nn.Linear(100, 1024),
nn.BatchNorm1d(1024),
nn.PReLU(),
nn.Linear(1024, 1024),
nn.BatchNorm1d(1024),
nn.PReLU(),
nn.Linear(1024, 1024),
nn.BatchNorm1d(1024),
nn.PReLU(),
nn.Linear(1024, 1024),
nn.BatchNorm1d(1024),
nn.PReLU(),
)
self.G1 = nn.Sequential(
nn.Linear(1024, 784),
nn.Tanh()
)
self.G2 = nn.Sequential(
nn.Linear(1024, 784),
nn.Tanh()
)
def forward(self, z):
z = z.view(z.size(0), z.size(1))
h = self.sw(z)
out_a = self.G1(h)
out_b = self.G2(h)
return out_a.view(z.size(0), 1, 28, 28), out_b.view(z.size(0), 1, 28, 28)
class R_CoGANDisMNIST(nn.Module):
def __init__(self):
super(R_CoGANDisMNIST, self).__init__()
## Domain A ##
self.conv0_a = nn.Conv2d(in_channels=1, out_channels=20, kernel_size=5, stride=1, padding=0)
self.pool0_a = nn.MaxPool2d(kernel_size=2)
self.conv1_a = nn.Conv2d(in_channels=20, out_channels=50, kernel_size=5, stride=1, padding=0)
self.pool1_a = nn.MaxPool2d(kernel_size=2)
## Domain B ##
self.conv0_b = nn.Conv2d(in_channels=1, out_channels=20, kernel_size=5, stride=1, padding=0)
self.pool0_b = nn.MaxPool2d(kernel_size=2)
self.conv1_b = nn.Conv2d(in_channels=20, out_channels=50, kernel_size=5, stride=1, padding=0)
self.pool1_b = nn.MaxPool2d(kernel_size=2)
self.shared_weight = nn.Sequential(
nn.Conv2d(in_channels=20, out_channels=50, kernel_size=5, stride=1, padding=0),
nn.MaxPool2d(kernel_size=2)
)
self.fc = nn.Sequential(
nn.Linear(800, 500),
nn.PReLU(),
)
########################
## Shared convolution ##
self.D = nn.Sequential(
nn.Linear(500,1),
nn.Sigmoid()
)
def forward(self, xa, xb):
h0_a = self.pool0_a(self.conv0_a(xa))
#h1_a = self.pool1_a(self.conv1_a(h0_a))
h0_b = self.pool0_b(self.conv0_b(xb))
#h1_b = self.pool1_b(self.conv1_b(h0_b))
#out_a = h0_a.view(h1_a.shape[0], -1)
#out_b = h0_a.view(h1_b.shape[0], -1)
o = torch.cat((h0_a, h0_b), 0)
h = self.shared_weight(o)
out_ = h.view(h.shape[0], -1)
out_a = self.fc(out_[:32])
out_b = self.fc(out_[32:])
out_ = torch.cat((out_a, out_b), 0)
out = self.D(out_)
out_a = out[:xa.shape[0]]
out_b = out[xa.shape[0]:]
return out_a, out_b
##################################################
## Conditional GAN
##################################################