-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmodels.py
164 lines (125 loc) · 7.37 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
import functools
import torch
import torch.nn as nn
def deconv_2d(in_channels,out_channels):
deconv=nn.ConvTranspose2d(in_channels, out_channels, kernel_size=2, stride=2)
return nn.Sequential(*[deconv])
class Blocks(nn.Module):
def __init__(self,res_block_num,conv_channels):
super(Blocks,self).__init__()
self.res_block_num=res_block_num
for res_block_idx in range(self.res_block_num):
conv_layer_1=nn.Conv2d(conv_channels,conv_channels,kernel_size=3,padding=1,stride=1)
conv_layer_2=nn.Conv2d(conv_channels,conv_channels,kernel_size=3,padding=1,stride=1)
self.add_module('%d'%(2*res_block_idx),nn.Sequential(conv_layer_1))
self.add_module('%d'%(2*res_block_idx+1),nn.Sequential(conv_layer_2))
def __getitem__(self, index):
if index < 0 or index >= len(self._modules):
raise IndexError('index %d is out of range'%(index))
return(self._modules[str(index)])
def __len__(self):
return self.res_block_num
class SeeMotionInDarkNet(nn.Module):
def __init__(self,input_channels=3,base_dim=32,res_block_num=16):
super(SeeMotionInDarkNet,self).__init__()
self.res_block_num = res_block_num
self.conv_1_1=nn.Conv2d(3,base_dim,kernel_size=3,padding=1,stride=1)
self.conv_1_2=nn.Conv2d(base_dim,base_dim,kernel_size=3,padding=1,stride=1)
self.conv_2_1=nn.Conv2d(base_dim,base_dim*2,kernel_size=3,padding=1,stride=1)
self.conv_2_2=nn.Conv2d(base_dim*2,base_dim*2,kernel_size=3,padding=1,stride=1)
self.conv_3_1=nn.Conv2d(base_dim*2,base_dim*4,kernel_size=3,padding=1,stride=1)
self.conv_3_2=nn.Conv2d(base_dim*4,base_dim*4,kernel_size=3,padding=1,stride=1)
self.conv_4_1=nn.Conv2d(base_dim*4,base_dim*8,kernel_size=3,padding=1,stride=1)
self.conv_4_2=nn.Conv2d(base_dim*8,base_dim*8,kernel_size=3,padding=1,stride=1)
self.conv_5_1=nn.Conv2d(base_dim*8,base_dim*16,kernel_size=3,padding=1,stride=1)
self.conv_5_2=nn.Conv2d(base_dim*16,base_dim*16,kernel_size=3,padding=1,stride=1)
self.res_block_list=Blocks(self.res_block_num,base_dim*16)
self.conv_after_res_block=nn.Conv2d(base_dim*16,base_dim*16,kernel_size=3,padding=1,stride=1)
self.deconv_6=deconv_2d(base_dim*16,base_dim*8)
self.conv_6_1=nn.Conv2d(base_dim*16,base_dim*8,kernel_size=3,padding=1,stride=1)
self.conv_6_2=nn.Conv2d(base_dim*8,base_dim*8,kernel_size=3,padding=1,stride=1)
self.deconv_7=deconv_2d(base_dim*8,base_dim*4)
self.conv_7_1=nn.Conv2d(base_dim*8,base_dim*4,kernel_size=3,padding=1,stride=1)
self.conv_7_2=nn.Conv2d(base_dim*4,base_dim*4,kernel_size=3,padding=1,stride=1)
self.deconv_8=deconv_2d(base_dim*4,base_dim*2)
self.conv_8_1=nn.Conv2d(base_dim*4,base_dim*2,kernel_size=3,padding=1,stride=1)
self.conv_8_2=nn.Conv2d(base_dim*2,base_dim*2,kernel_size=3,padding=1,stride=1)
self.deconv_9=deconv_2d(base_dim*2,base_dim)
self.conv_9_1=nn.Conv2d(base_dim*2,base_dim,kernel_size=3,padding=1,stride=1)
self.conv_9_2=nn.Conv2d(base_dim,base_dim,kernel_size=3,padding=1,stride=1)
self.conv_10=nn.Conv2d(base_dim,3,kernel_size=3,padding=1,stride=1)
def forward(self, input_tensor):
self.pool=nn.MaxPool2d(2)
self.lrelu=nn.LeakyReLU(negative_slope=0.2)
conv_1=self.lrelu(self.conv_1_1(input_tensor))
conv_1=self.conv_1_2(conv_1)
self.ln1 = nn.LayerNorm(conv_1.size()[1:], elementwise_affine=False)
conv_1 = self.lrelu(self.ln1(conv_1))
conv_1_pool=self.pool(conv_1)
conv_2=self.conv_2_1(conv_1_pool)
self.ln2_1 = nn.LayerNorm(conv_2.size()[1:], elementwise_affine=False)
conv_2=self.conv_2_2(self.lrelu(self.ln2_1(conv_2)))
self.ln2_2 = nn.LayerNorm(conv_2.size()[1:], elementwise_affine=False)
conv_2=self.lrelu(self.ln2_2(conv_2))
conv_2_pool=self.pool(conv_2)
conv_3=self.conv_3_1(conv_2_pool)
self.ln3_1 = nn.LayerNorm(conv_3.size()[1:], elementwise_affine=False)
conv_3=self.conv_3_2(self.lrelu(self.ln3_1(conv_3)))
self.ln3_2 = nn.LayerNorm(conv_3.size()[1:], elementwise_affine=False)
conv_3=self.lrelu(self.ln3_2(conv_3))
conv_3_pool=self.pool(conv_3)
conv_4=self.conv_4_1(conv_3_pool)
self.ln4_1 = nn.LayerNorm(conv_4.size()[1:], elementwise_affine=False)
conv_4=self.conv_4_2(self.lrelu(self.ln4_1(conv_4)))
self.ln4_2 = nn.LayerNorm(conv_4.size()[1:], elementwise_affine=False)
conv_4=self.lrelu(self.ln4_2(conv_4))
conv_4_pool=self.pool(conv_4)
conv_5=self.conv_5_1(conv_4_pool)
self.ln5_1 = nn.LayerNorm(conv_5.size()[1:], elementwise_affine=False)
conv_5=self.conv_5_2(self.lrelu(self.ln5_1(conv_5)))
self.ln5_2 = nn.LayerNorm(conv_5.size()[1:], elementwise_affine=False)
conv_5=self.lrelu(self.ln5_2(conv_5))
#res block
conv_feature_end=conv_5
for res_block_idx in range(self.res_block_num):
conv_feature_begin = conv_feature_end
conv_feature=self.res_block_list[2*res_block_idx](conv_feature_begin)
self.ln_1 = nn.LayerNorm(conv_feature.size()[1:], elementwise_affine=False)
conv_feature = self.res_block_list[2*res_block_idx+1](self.lrelu(self.ln_1(conv_feature)))
self.ln_2 = nn.LayerNorm(conv_feature.size()[1:], elementwise_affine=False)
conv_feature = self.ln_2(conv_feature)
conv_feature_end = conv_feature_begin+conv_feature
conv_feature=self.conv_after_res_block(conv_feature_end)
self.ln = nn.LayerNorm(conv_feature.size()[1:], elementwise_affine=False)
conv_feature = self.ln(conv_feature)
conv_feature=conv_feature+conv_5
conv_6_up=self.deconv_6(conv_feature)
conv_6=torch.cat((conv_6_up,conv_4),dim=1)
conv_6=self.conv_6_1(conv_6)
self.ln6_1 = nn.LayerNorm(conv_6.size()[1:], elementwise_affine=False)
conv_6=self.conv_6_2(self.lrelu(self.ln6_1(conv_6)))
self.ln6_2 = nn.LayerNorm(conv_6.size()[1:], elementwise_affine=False)
conv_6=self.lrelu(self.ln6_2(conv_6))
conv_7_up=self.deconv_7(conv_6)
conv_7=torch.cat((conv_7_up,conv_3),dim=1)
conv_7=self.conv_7_1(conv_7)
self.ln7_1 = nn.LayerNorm(conv_7.size()[1:], elementwise_affine=False)
conv_7=self.conv_7_2(self.lrelu(self.ln7_1(conv_7)))
self.ln7_2 = nn.LayerNorm(conv_7.size()[1:], elementwise_affine=False)
conv_7=self.lrelu(self.ln7_2(conv_7))
conv_8_up=self.deconv_8(conv_7)
conv_8=torch.cat((conv_8_up,conv_2),dim=1)
conv_8=self.conv_8_1(conv_8)
self.ln8_1 = nn.LayerNorm(conv_8.size()[1:], elementwise_affine=False)
conv_8=self.conv_8_2(self.lrelu(self.ln8_1(conv_8)))
self.ln8_2 = nn.LayerNorm(conv_8.size()[1:], elementwise_affine=False)
conv_8=self.lrelu(self.ln8_2(conv_8))
conv_9_up=self.deconv_9(conv_8)
conv_9=torch.cat((conv_9_up,conv_1),dim=1)
conv_9=self.conv_9_1(conv_9)
self.ln9_1 = nn.LayerNorm(conv_9.size()[1:], elementwise_affine=False)
conv_9=self.conv_9_2(self.lrelu(self.ln9_1(conv_9)))
self.ln9_2 = nn.LayerNorm(conv_9.size()[1:], elementwise_affine=False)
conv_9=self.lrelu(self.ln9_2(conv_9))
out=self.conv_10(conv_9)
return out