-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodels.py
185 lines (150 loc) · 6.49 KB
/
models.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
import torch
import torch.nn as nn
import numpy as np
import math
class UpUint(nn.Module):
def __init__(self,num_input_features,num_output_features,kernel_size,stride,padding):
super(UpUint,self).__init__()
self.deconv1 = nn.ConvTranspose2d(in_channels=num_input_features, out_channels=num_output_features,kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu1 = nn.PReLU()
self.conv1 = nn.Conv2d(in_channels=num_output_features, out_channels=num_input_features, kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu2 = nn.PReLU()
self.deconv2 = nn.ConvTranspose2d(in_channels=num_input_features, out_channels=num_output_features,kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu3 = nn.PReLU()
def forward(self,x):
h0 = self.relu1(self.deconv1(x))
l0 = self.relu2(self.conv1(h0))
diff = l0 - x
h1 = self.relu3(self.deconv2(diff))
out = h1 + h0
return out
class DownUint(nn.Module):
def __init__(self,num_input_features,num_output_features,kernel_size,stride,padding):
super(DownUint,self).__init__()
self.conv1 = nn.Conv2d(in_channels=num_input_features, out_channels=num_output_features, kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu1 = nn.PReLU()
self.deconv1 = nn.ConvTranspose2d(in_channels=num_output_features, out_channels=num_input_features,kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu2 = nn.PReLU()
self.conv2 = nn.Conv2d(in_channels=num_input_features, out_channels=num_output_features, kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
self.relu3 = nn.PReLU()
def forward(self,x):
l0 = self.relu1(self.conv1(x))
h0 = self.relu2(self.deconv1(l0))
diff = h0 - x
l1 = self.relu3(self.conv2(diff))
out = l1 + l0
return out
class TranSition(nn.Module):
def __init__(self,num_input_features,num_output_features):
super(TranSition,self).__init__()
self.conv1 = nn.Conv2d(in_channels=num_input_features, out_channels=num_output_features, kernel_size=1, stride=1, padding=0, bias=False)
self.relu1 = nn.PReLU()
def forward(self,x):
out = self.relu1(self.conv1(x))
return out
class Net(nn.Module):
def __init__(self,scale_factor):
super(Net,self).__init__()
if scale_factor == 2:
self.kernel_size = 6
self.stride = 2
self.padding = 2
elif scale_factor == 4:
self.kernel_size = 8
self.stride = 4
self.padding = 2
elif scale_factor == 8:
self.kernel_size = 12
self.stride = 8
self.padding = 2
self.conv1 = nn.Conv2d(3,256,kernel_size=3,stride=1,padding=1,bias=False)
self.relu1 = nn.PReLU()
self.conv2 = nn.Conv2d(256,64,kernel_size=1,stride=1,padding=0,bias=False)
self.relu2 = nn.PReLU()
self.up1 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.down1 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans1 = TranSition(128,64)
self.up2 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans2 = TranSition(128,64)
self.down2 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans22 = TranSition(192,64)
self.up3 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans3 = TranSition(64*3,64)
self.down3 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans32 = TranSition(64*4,64)
self.up4 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans4 = TranSition(64*4,64)
self.down4 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans42 = TranSition(64*5,64)
self.up5 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans5 = TranSition(64*5,64)
self.down5 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans52 = TranSition(64*6,64)
self.up6 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans6 = TranSition(64*6,64)
self.down6 = DownUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans62 = TranSition(64*7,64)
self.up7 = UpUint(64,64,self.kernel_size,self.stride,self.padding)
self.trans71 = TranSition(64*7,64)
self.reconv = nn.Conv2d(64,3,kernel_size=3,stride=1,padding=1,bias=False)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.ConvTranspose2d):
torch.nn.init.kaiming_normal(m.weight)
if m.bias is not None:
m.bias.data.zero_()
def forward(self,x):
features = self.relu2(self.conv2(self.relu1(self.conv1(x))))
h1 = self.up1(features)
l1 = self.down1(h1)
lin = torch.cat((features,l1),1)
lin11 = self.trans1(lin)
h2 = self.up2(lin11)
hin = torch.cat((h1,h2),1)
hin11 = self.trans2(hin)
l2 = self.down2(hin11)
lin2 = torch.cat((lin,l2),1)
lin21 = self.trans22(lin2)
h3 = self.up3(lin21)
hin2 = torch.cat((hin,h3),1)
hin21 = self.trans3(hin2)
l3 = self.down3(hin21)
lin3 = torch.cat((lin2,l3),1)
lin31 = self.trans32(lin3)
h4 = self.up4(lin31)
hin3 = torch.cat((hin2,h4),1)
hin31 = self.trans4(hin3)
l4 = self.down4(hin31)
lin4 = torch.cat((lin3,l4),1)
lin41 = self.trans42(lin4)
h5 = self.up5(lin41)
hin4 = torch.cat((hin3,h5),1)
hin41 = self.trans5(hin4)
l5 = self.down5(hin41)
lin5 = torch.cat((lin4,l5),1)
lin51 = self.trans52(lin5)
h6 = self.up6(lin51)
hin5 = torch.cat((hin4,h6),1)
hin51 = self.trans6(hin5)
l6 = self.down6(hin51)
lin6 = torch.cat((lin5,l6),1)
lin61 = self.trans62(lin6)
h7 = self.up7(lin61)
hout = torch.cat((hin5,h7),1)
hout = self.trans71(hout)
out = self.reconv(hout)
return out
class L1_Charbonnier_loss(nn.Module):
"""L1 Charbonnierloss."""
def __init__(self):
super(L1_Charbonnier_loss, self).__init__()
self.eps = 1e-6
def forward(self, X, Y):
diff = torch.add(X, -Y)
error = torch.sqrt( diff * diff + self.eps )
loss = torch.mean(error)
return loss