-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathAEGeAN_512.py
189 lines (171 loc) · 6.92 KB
/
AEGeAN_512.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
import torch
from torch.autograd import Variable
import torch.nn as nn
from collections import OrderedDict as odict
class GAN_G(nn.Module):
def __init__(self, nz, ngf, nc=3):
super(GAN_G, self).__init__()
self.main = nn.Sequential()
self.layers = odict([
# block 0
('convT_0', nn.ConvTranspose2d(nz, ngf * \
32, 4, stride=2, padding=1, bias=False)),
('bn_0', nn.BatchNorm2d(ngf * 32)),
('act_0', nn.LeakyReLU(inplace=True)),
# block 1
('convT_1', nn.ConvTranspose2d(ngf * 32, ngf * \
16, 4, stride=2, padding=1, bias=False)),
('bn_1', nn.BatchNorm2d(ngf * 16)),
('act_1', nn.LeakyReLU(inplace=True)),
# block 2
('convT_2', nn.ConvTranspose2d(ngf * 16, ngf * \
16, 4, stride=2, padding=1, bias=False)),
('bn_2', nn.BatchNorm2d(ngf * 16)),
('act_2', nn.LeakyReLU(inplace=True)),
# block 3
('convT_3', nn.ConvTranspose2d(ngf * 16, ngf * \
16, 4, stride=2, padding=1, bias=False)),
('bn_3', nn.BatchNorm2d(ngf * 16)),
('act_3', nn.LeakyReLU(inplace=True)),
# block 4
('convT_4', nn.ConvTranspose2d(ngf * 16, ngf * \
32, 4, stride=2, padding=1, bias=False)),
('bn_4', nn.BatchNorm2d(ngf * 32)),
('act_4', nn.LeakyReLU(inplace=True)),
# block 5
('convT_5', nn.ConvTranspose2d(ngf * 32, ngf * \
16, 4, stride=2, padding=1, bias=False)),
('bn_5', nn.BatchNorm2d(ngf * 16)),
('act_5', nn.LeakyReLU(inplace=True)),
# block 6
('convT_6', nn.ConvTranspose2d(ngf * 16, ngf * \
16, 4, stride=2, padding=1, bias=False)),
('bn_6', nn.BatchNorm2d(ngf * 16)),
('act_6', nn.LeakyReLU(inplace=True)),
# block 7
('convT_7', nn.ConvTranspose2d(ngf * 16,
ngf * 8, 4, stride=2, padding=1, bias=False)),
('bn_7', nn.BatchNorm2d(ngf * 8)),
('act_7', nn.LeakyReLU(inplace=True)),
# block 8
('convT_8', nn.ConvTranspose2d(
ngf * 8, nc, 4, stride=2, padding=1, bias=False)),
('act_8', nn.LeakyReLU(inplace=True)),
])
self.tanh = nn.Tanh()
self.sigmoid = nn.Sigmoid()
def build(self):
for name, layer in self.layers.items():
self.main.add_module(name, layer)
def forward(self, x, mode='gen'):
out = self.main(x)
if mode == 'gen':
out = self.tanh(out)
if mode == 'ae':
out = self.sigmoid(out)
return out
def n_params(self):
params = [[num for num in param.size()]
for param in self.main.parameters()]
n = 0
param_graph = []
for group in params:
s = 1
for item in group:
s *= item
param_graph += [s]
n += s
return n, param_graph
class GAN_D(nn.Module):
def __init__(self, nc, ndf, nout):
super(GAN_D, self).__init__()
self.nout = nout
self.main = nn.Sequential()
self.layers = odict([
# block 0
# input: (batch x nc x 512 x 512)
('conv_1', nn.Conv2d(nc, ndf * 8, 4, stride=2, padding=1, bias=False)),
('bn_1', nn.BatchNorm2d(ndf * 8)),
('act_1', nn.LeakyReLU(0.2, inplace=True)),
# block 2
# input: (batch x nc x 256 x 256)
('conv_2', nn.Conv2d(ndf * 8, ndf * 16,
4, stride=2, padding=1, bias=False)),
('bn_2', nn.BatchNorm2d(ndf * 16)),
('act_2', nn.LeakyReLU(0.2, inplace=True)),
# block 3
# input: (batch x nc x 128 x 128)
('conv_3', nn.Conv2d(ndf * 16, ndf * 16,
4, stride=2, padding=1, bias=False)),
('bn_3', nn.BatchNorm2d(ndf * 16)),
('act_3', nn.LeakyReLU(0.2, inplace=True)),
# block 4
# input: (batch x nc x 64 x 64)
('conv_4', nn.Conv2d(ndf * 16, ndf * 16,
4, stride=2, padding=1, bias=False)),
('bn_4', nn.BatchNorm2d(ndf * 16)),
('act_4', nn.LeakyReLU(0.2, inplace=True)),
# block 5
# input: (batch x nc x 32 x 32)
('conv_5', nn.Conv2d(ndf * 16, ndf * 16,
4, stride=2, padding=1, bias=False)),
('bn_5', nn.BatchNorm2d(ndf * 16)),
('act_5', nn.LeakyReLU(0.2, inplace=True)),
# block 6
# input: (batch x nc x 16 x 16)
('conv_6', nn.Conv2d(ndf * 16, ndf * 32,
4, stride=2, padding=1, bias=False)),
('bn_6', nn.BatchNorm2d(ndf * 32)),
('act_6', nn.LeakyReLU(0.2, inplace=True)),
# block 7
# input: (batch x nc x 8 x 8)
('conv_7', nn.Conv2d(ndf * 32, ndf * 32,
4, stride=2, padding=1, bias=False)),
('bn_7', nn.BatchNorm2d(ndf * 32)),
('act_7', nn.LeakyReLU(0.2, inplace=True)),
# block 8
# input: (batch x nc x 4 x 4)
('conv_8', nn.Conv2d(ndf * 32, ndf * 32,
4, stride=2, padding=1, bias=False)),
('bn_8', nn.BatchNorm2d(ndf * 32)),
('act_8', nn.LeakyReLU()),
# block9
# input: (batch x nc x 2 x 2)
('conv_9', nn.Conv2d(ndf * 32, nout, 2, stride=2, padding=0, bias=False)),
('act_9', nn.LeakyReLU()),
])
def build(self):
for name, layer in self.layers.items():
self.main.add_module(name, layer)
def forward(self, x):
out = self.main(x)
return out.view(-1, self.nout)
def n_params(self):
params = [[num for num in param.size()]
for param in self.main.parameters()]
n = 0
param_graph = []
for group in params:
s = 1
for item in group:
s *= item
param_graph += [s]
n += s
return n, param_graph
if __name__ == '__main__':
import matplotlib.pyplot as plt
x = torch.randn(1, 100, 1, 1)
net = GAN_G(100, 24, 3)
net.build()
np, pg = net.n_params()
print('{:,}'.format(np))
# net.cuda()
out = net(x)
print(out.size())
xs = range(len(pg))
plt.bar(xs, pg, color='k')
plt.yscale('log')
for xloc, v in zip(xs, pg):
plt.text(xloc, max(pg) / 2,
s='{:,}'.format(v), color=(1, 0, 1), rotation=90)
plt.show()