-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathmodel.py
207 lines (157 loc) · 6.83 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
200
201
202
203
204
205
206
207
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision.models as models
class Generator(nn.Module):
def __init__(self):
super(Generator, self).__init__()
haze_class = models.densenet121(pretrained=True)
############# Block0-down ##############
self.conv0 = haze_class.features.conv0
self.relu0 = haze_class.features.relu0
self.pool0 = haze_class.features.pool0
############# Block1-down ##############
self.dense_block1 = haze_class.features.denseblock1
self.trans_block1 = haze_class.features.transition1
############# Block2-down ##############
self.dense_block2 = haze_class.features.denseblock2
self.trans_block2 = haze_class.features.transition2
############# Block3-down ##############
self.dense_block3 = haze_class.features.denseblock3
self.trans_block3 = haze_class.features.transition3
self.res31 = BasicResBlock(512, 512)
self.res32 = BasicResBlock(512, 512)
############# Block4-up ##############
self.dense_block4 = BottleneckBlock(512, 256)
self.trans_block4 = TransitionBlock(768, 128)
self.res41 = BasicResBlock(387, 387)
self.res42 = BasicResBlock(387, 387)
############# Block3-up ##############
self.dense_block5 = BottleneckBlock(387, 256)
self.trans_block5 = TransitionBlock(643, 128)
self.res51 = BasicResBlock(259, 259)
self.res52 = BasicResBlock(259, 259)
############# Block2-up ##############
self.dense_block6 = BottleneckBlock(259, 128)
self.trans_block6 = TransitionBlock(387, 64)
self.res61 = BasicResBlock(67, 67)
self.res62 = BasicResBlock(67, 67)
############# Block1-up ##############
self.dense_block7 = BottleneckBlock(67, 64)
self.trans_block7 = TransitionBlock(131, 32)
self.res71 = BasicResBlock(35, 35)
self.res72 = BasicResBlock(35, 35)
############# Block0-up ##############
self.dense_block8 = BottleneckBlock(35, 32)
self.trans_block8 = TransitionBlock(67, 16)
# multi-scale
self.conv_refin = nn.Conv2d(19, 20, 3, 1, 1)
self.refout = nn.Conv2d(20, 3, kernel_size=3, stride=1, padding=1)
self.relu = nn.LeakyReLU(0.2, inplace=True)
self.tanh = nn.Tanh()
def forward(self, x):
## input: 1024 x 1024
x0 = self.pool0(self.relu0(self.conv0(x))) # 256 x 256
x1 = self.dense_block1(x0) # 256 x 256
x1 = self.trans_block1(x1) # 128 x 128
x2 = self.trans_block2(self.dense_block2(x1)) # 64 x 64
x3 = self.trans_block3(self.dense_block3(x2)) # 32 x 32
x3 = self.res31(x3) # 32 x 32
x3 = self.res32(x3) # 32 x 32
x4 = self.trans_block4(self.dense_block4(x3)) # 64 x 64
x43 = F.avg_pool2d(x, 16) # 64 x 64
x42 = torch.cat([x4, x2, x43], 1) # 64 x 64
x42 = self.res41(x42) # 64
x42 = self.res42(x42) # 64
x5 = self.trans_block5(self.dense_block5(x42)) # 128
x53 = F.avg_pool2d(x, 8) # 128
x52 = torch.cat([x5, x1, x53], 1) # 128
x52 = self.res51(x52) # 128
x52 = self.res52(x52) # 128
x6 = self.trans_block6(self.dense_block6(x52)) # 256
x63 = F.avg_pool2d(x, 4) # 256
x62 = torch.cat([x6, x63], 1) # 256
x62 = self.res61(x62) # 256
x6 = self.res62(x62) # 256
x7 = self.trans_block7(self.dense_block7(x6)) # 512
x73 = F.avg_pool2d(x, 2) # 512
x72 = torch.cat([x7, x73], 1) # 512
x72 = self.res71(x72) # 512
x7 = self.res72(x72) # 512
x8 = self.trans_block8(self.dense_block8(x7)) # 1024
x8 = torch.cat([x8, x], 1) # 1024
x9 = self.relu(self.conv_refin(x8)) # 1024
dehaze = self.tanh(self.refout(x9))
return dehaze
class BottleneckBlock(nn.Module):
def __init__(self, in_planes, out_planes):
super(BottleneckBlock, self).__init__()
inter_planes = out_planes * 4
self.relu = nn.ReLU()
self.conv1 = nn.Conv2d(in_planes, inter_planes, kernel_size=1, stride=1,
padding=0, bias=False)
self.conv2 = nn.Conv2d(inter_planes, out_planes, kernel_size=3, stride=1,
padding=1, bias=False)
def forward(self, x):
out = self.conv1(self.relu(x))
out = self.conv2(self.relu(out))
return torch.cat([x, out], 1)
class TransitionBlock(nn.Module):
def __init__(self, in_planes, out_planes):
super(TransitionBlock, self).__init__()
self.relu = nn.ReLU()
self.conv1 = nn.ConvTranspose2d(in_planes, out_planes, kernel_size=1, stride=1,
padding=0, bias=False)
def forward(self, x):
out = self.conv1(self.relu(x))
return F.interpolate(out, scale_factor=2)
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
padding=1, bias=False)
class BasicResBlock(nn.Module):
def __init__(self, inplanes, planes):
super(BasicResBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.relu(out)
out = self.conv2(out)
out += residual
out = self.relu(out)
return out
class Discriminator(nn.Module):
def __init__(self, nc=3, nf=36):
super(Discriminator, self).__init__()
self.main = nn.Sequential(
nn.Conv2d(nc, nf, kernel_size=4, stride=2, padding=1, bias=False), # 36 x 512 x 512
nn.LeakyReLU(0.2, inplace=True),
DBlock(nf, nf * 2), # 72 x 256 x 256
DBlock(nf * 2, nf * 4), # 144 x 128 x 128
DBlock(nf * 4, nf * 8), # 288 x 64 x 64
DBlock(nf * 8, nf * 8), # 288 x 32 x 632
nn.Conv2d(nf * 8, nf * 8, 4, 1, 1, bias=False), # 288 x 31 x 31
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(nf * 8, 1, 4, 1, 1, bias=False), # 288 x 30 x 30
nn.Sigmoid()
)
def forward(self, x):
output = self.main(x)
return output
def requires_grad(self, req):
for param in self.parameters():
param.requires_grad = req
class DBlock(nn.Module):
def __init__(self, in_c, out_c):
super(DBlock, self).__init__()
self.main = nn.Sequential(
nn.Conv2d(in_c, out_c, kernel_size=3, stride=2, padding=1, bias=False),
nn.BatchNorm2d(out_c),
nn.LeakyReLU(0.2, inplace=True)
)
def forward(self, x):
output = self.main(x)
return output