-
Notifications
You must be signed in to change notification settings - Fork 1
/
nets_YK.py
175 lines (144 loc) · 6.64 KB
/
nets_YK.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
import torch
import torch.nn.functional as F
#from torchsummary import summary
from torch import nn
from modules import DilatedResidualBlock, NLB, MTUR, DepthWiseDilatedResidualBlock
class MTURNet_YKCCR(nn.Module):
def __init__(self, num_features=64):
super(MTURNet_YKCCR, self).__init__()
self.mean = torch.zeros(1, 3, 1, 1)
self.std = torch.zeros(1, 3, 1, 1)
self.mean[0, 0, 0, 0] = 0.485
self.mean[0, 1, 0, 0] = 0.456
self.mean[0, 2, 0, 0] = 0.406
self.std[0, 0, 0, 0] = 0.229
self.std[0, 1, 0, 0] = 0.224
self.std[0, 2, 0, 0] = 0.225
self.mean = nn.Parameter(self.mean)
self.std = nn.Parameter(self.std)
self.mean.requires_grad = False
self.std.requires_grad = False
############################################ MT prediction network
self.conv1 = nn.Sequential(
nn.Conv2d(3, 32, 4, stride=2, padding=1),
nn.GroupNorm(num_groups=32, num_channels=32),
nn.SELU(inplace=True)
)
self.conv2 = nn.Sequential(
nn.Conv2d(64, 64, 4, stride=2, padding=1),
nn.GroupNorm(num_groups=32, num_channels=64),
nn.SELU(inplace=True)
)
self.conv10 = nn.Sequential(
nn.ConvTranspose2d(64, 64, kernel_size=4, stride=2, padding=1),
nn.GroupNorm(num_groups=32, num_channels=64),
nn.SELU(inplace=True)
)
self.conv1_1 = nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0)
self.depth_pred = nn.Sequential(
nn.ConvTranspose2d(64, 32, kernel_size=4, stride=2, padding=1),
nn.GroupNorm(num_groups=32, num_channels=32),
nn.SELU(inplace=True),
nn.Conv2d(32, 32, kernel_size=3, padding=1),
nn.SELU(inplace=True),
nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
# nn.Sigmoid()
)
self.f_process = nn.Sequential(
nn.ConvTranspose2d(64, 64, kernel_size=4, stride=2, padding=1),
nn.GroupNorm(num_groups=64, num_channels=64),
nn.SELU(inplace=True),
)
self.depth_process = nn.Sequential(
nn.Conv2d(65, 128, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
# nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
# nn.Sigmoid()
)
############################################ underwater enhanced network
self.head = nn.Sequential(
nn.Conv2d(3, 32, kernel_size=4, stride=2, padding=1), nn.ReLU(),
nn.Conv2d(32, num_features, kernel_size=1, stride=1, padding=0), nn.ReLU()
)
self.body1 = nn.Sequential(
DilatedResidualBlock(num_features, 1),)
self.body2 = nn.Sequential(
DilatedResidualBlock(num_features, 2))
self.body4 = nn.Sequential(
DilatedResidualBlock(num_features, 4))
self.body8 = nn.Sequential(
DilatedResidualBlock(num_features, 8))
self.mturb = MTUR(num_features)
self.tail = nn.Sequential(
# nn.Conv2d(num_features, num_features, kernel_size=3, padding=1), nn.ReLU(),
nn.ConvTranspose2d(num_features, 32, kernel_size=4, stride=2, padding=1), nn.ReLU(),
nn.Conv2d(32, 3, kernel_size=3, padding=1)
)
self.c3_31 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_32 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_33 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_34 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_35 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_36 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_37 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_38 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),nn.ReLU(inplace=True))
self.c3_39 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1), nn.ReLU(inplace=True))
self.c3_310 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1), nn.ReLU(inplace=True))
self.output = nn.Sequential(
nn.Conv2d(128, 64, kernel_size=3, padding=1), nn.ReLU(),
# nn.ConvTranspose2d(num_features, 32, kernel_size=4, stride=2, padding=1), nn.ReLU(),
nn.Conv2d(64, 3, kernel_size=3, padding=1)
)
for m in self.modules():
if isinstance(m, nn.ReLU):
m.inplace = True
def forward(self, x):
x = (x - self.mean) / self.std
f = self.head(x)
################################## MT prediction network
d_f2 = self.conv2(f) # torch.Size([8, 64, 32, 32])
d_f10 = self.conv10(d_f2) # torch.Size([8, 64, 64, 64])
depth_pred = self.depth_pred(d_f10) # torch.Size([8, 1, 128, 128])
################################## underwater enhanced network
f = self.body1(f) # torch.Size([8, 64, 64, 64])
c = self.c3_31(d_f10)
f = self.body1(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_32(d_f10)
f = self.body2(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_33(d_f10)
f = self.body2(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_34(d_f10)
f = self.body4(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_35(d_f10)
f = self.body8(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_36(d_f10)
f = self.body4(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_37(d_f10)
f = self.body2(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_38(d_f10)
f = self.body2(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_39(d_f10)
f = self.body1(f+c) # torch.Size([8, 64, 64, 64])
c = self.c3_310(d_f10)
f = self.body1(f+c) # torch.Size([8, 64, 64, 64])
f = self.f_process(f) # f_process(f) torch.Size([8, 64, 128, 128])
f = torch.cat((f, depth_pred.detach()), 1) # torch.Size([8, 65, 128, 128])
f = self.depth_process(f) # torch.Size([8, 128, 128, 128])
x = self.output(f)
x = (x * self.std + self.mean).clamp(min=0, max=1)
# print("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~")
# print('x', x.shape)
# input()
if self.training:
return x, depth_pred
return x