/
DeblurNet.py
184 lines (155 loc) · 6.67 KB
/
DeblurNet.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
#!/usr/bin/python
# -*- coding: utf-8 -*-
#
# Developed by Shangchen Zhou <shangchenzhou@gmail.com>
from models.submodules import *
from models.FAC.kernelconv2d import KernelConv2D
from torch import nn
# # model size
# class DeblurNet(nn.Module):
# def __init__(self):
# super(DeblurNet, self).__init__()
# # OVD size
# self.conv1 = conv(5*3, 64, kernel_size=5, stride=1)
# self.conv2 = conv(64, 32, kernel_size=3, stride=2)
# self.conv3 = resnet_block(64, kernel_size=3)
# self.conv4 = resnet_block(64, kernel_size=3)
# self.conv5 = resnet_block(64, kernel_size=3)
# self.conv6 = resnet_block(64, kernel_size=3)
#
# self.conv7 = conv(128, 64, kernel_size=5, stride=1)
#
# self.conv8 = resnet_block(64, kernel_size=3)
# self.conv9 = resnet_block(64, kernel_size=3)
# self.conv10 = resnet_block(64, kernel_size=3)
# self.conv11 = resnet_block(64, kernel_size=3)
# self.conv12 = upconv(64, 64)
# self.conv13 = conv(64, 3,kernel_size=3)
#
# self.conv14 = conv(64, 32,kernel_size=3)
#
# # su size
# # self.conv1 = conv(15,64,kernel_size=5)
# #
# # self.conv2 = conv(64,64,kernel_size=3)
# # self.conv3 = conv(64,128,kernel_size=3)
# # self.conv4 = conv(128,128,kernel_size=3)
# #
# # self.conv5 = conv(128,256,kernel_size=3)
# # self.conv6 = conv(256,256,kernel_size=3)
# # self.conv7 = conv(256,256,kernel_size=3)
# # self.conv8 = conv(256,256,kernel_size=3)
# #
# # self.conv9 = conv(256,512,kernel_size=3)
# # self.conv10 = conv(512,512,kernel_size=3)
# # self.conv11 = conv(512,512,kernel_size=3)
# # self.conv12 = conv(512,512,kernel_size=3)
# #
# # self.conv13 = upconv(768,256)
# # self.conv14 = conv(256,256,kernel_size=3)
# # self.conv15 = conv(256,256,kernel_size=3)
# # self.conv16 = conv(256,256,kernel_size=3)
# #
# # self.conv17 = upconv(384,128)
# # self.conv18 = conv(128,128,kernel_size=3)
# # self.conv19 = conv(128,64,kernel_size=3)
# #
# # self.conv20 = upconv(128, 64)
# # self.conv21 = conv(64, 15, kernel_size=3)
# # self.conv22 = conv(15, 3, kernel_size=3)
#
# def forward(self, img_blur, output_last_img_clear):
# pass
# return 1
class DeblurNet(nn.Module):
def __init__(self):
super(DeblurNet, self).__init__()
#############################
# Deblurring Branch
#############################
# encoder
ks = 3
ks_2d = 5
ch1 = 32
ch2 = 64
ch3 = 128
self.fea = conv(2*ch3, ch3, kernel_size=ks, stride=1)
self.conv1_1 = conv(3, ch1, kernel_size=ks, stride=1)
self.conv1_2 = resnet_block(ch1, kernel_size=ks)
self.conv1_3 = resnet_block(ch1, kernel_size=ks)
self.conv2_1 = conv(ch1, ch2, kernel_size=ks, stride=2)
self.conv2_2 = resnet_block(ch2, kernel_size=ks)
self.conv2_3 = resnet_block(ch2, kernel_size=ks)
self.conv3_1 = conv(ch2, ch3, kernel_size=ks, stride=2)
self.conv3_2 = resnet_block(ch3, kernel_size=ks)
self.conv3_3 = resnet_block(ch3, kernel_size=ks)
self.kconv_warp = KernelConv2D.KernelConv2D(kernel_size=ks_2d)
self.kconv_deblur = KernelConv2D.KernelConv2D(kernel_size=ks_2d)
# decoder
self.upconv2_u = upconv(2*ch3, ch2)
self.upconv2_2 = resnet_block(ch2, kernel_size=ks)
self.upconv2_1 = resnet_block(ch2, kernel_size=ks)
self.upconv1_u = upconv(ch2, ch1)
self.upconv1_2 = resnet_block(ch1, kernel_size=ks)
self.upconv1_1 = resnet_block(ch1, kernel_size=ks)
self.img_prd = conv(ch1, 3, kernel_size=ks)
#############################
# Kernel Prediction Branch
#############################
# kernel network
self.kconv1_1 = conv(9, ch1, kernel_size=ks, stride=1)
self.kconv1_2 = resnet_block(ch1, kernel_size=ks)
self.kconv1_3 = resnet_block(ch1, kernel_size=ks)
self.kconv2_1 = conv(ch1, ch2, kernel_size=ks, stride=2)
self.kconv2_2 = resnet_block(ch2, kernel_size=ks)
self.kconv2_3 = resnet_block(ch2, kernel_size=ks)
self.kconv3_1 = conv(ch2, ch3, kernel_size=ks, stride=2)
self.kconv3_2 = resnet_block(ch3, kernel_size=ks)
self.kconv3_3 = resnet_block(ch3, kernel_size=ks)
self.fac_warp = nn.Sequential(
conv(ch3, ch3, kernel_size=ks),
resnet_block(ch3, kernel_size=ks),
resnet_block(ch3, kernel_size=ks),
conv(ch3, ch3 * ks_2d ** 2, kernel_size=1))
self.kconv4 = conv(ch3 * ks_2d ** 2, ch3, kernel_size=1)
self.fac_deblur = nn.Sequential(
conv(2*ch3, ch3, kernel_size=ks),
resnet_block(ch3, kernel_size=ks),
resnet_block(ch3, kernel_size=ks),
conv(ch3, ch3 * ks_2d ** 2, kernel_size=1))
def forward(self, img_blur, last_img_blur, output_last_img, output_last_fea):
merge = torch.cat([img_blur, last_img_blur, output_last_img], 1)
#############################
# Kernel Prediction Branch
#############################
# kernel network
kconv1 = self.kconv1_3(self.kconv1_2(self.kconv1_1(merge)))
kconv2 = self.kconv2_3(self.kconv2_2(self.kconv2_1(kconv1)))
kconv3 = self.kconv3_3(self.kconv3_2(self.kconv3_1(kconv2)))
# fac
kernel_warp = self.fac_warp(kconv3)
kconv4 = self.kconv4(kernel_warp)
kernel_deblur = self.fac_deblur(torch.cat([kconv3, kconv4],1))
#############################
# Deblurring Branch
#############################
# encoder blur
conv1_d = self.conv1_1(img_blur)
conv1_d = self.conv1_3(self.conv1_2(conv1_d))
conv2_d = self.conv2_1(conv1_d)
conv2_d = self.conv2_3(self.conv2_2(conv2_d))
conv3_d = self.conv3_1(conv2_d)
conv3_d = self.conv3_3(self.conv3_2(conv3_d))
conv3_d_k = self.kconv_deblur(conv3_d, kernel_deblur)
# encoder last_clear
if output_last_fea is None:
output_last_fea = torch.cat([conv3_d, conv3_d],1)
output_last_fea = self.fea(output_last_fea)
conv_a_k = self.kconv_deblur(output_last_fea, kernel_warp)
conv3 = torch.cat([conv3_d_k, conv_a_k],1)
# decoder
upconv2 = self.upconv2_1(self.upconv2_2(self.upconv2_u(conv3)))
upconv1 = self.upconv1_1(self.upconv1_2(self.upconv1_u(upconv2)))
output_img = self.img_prd(upconv1) + img_blur
output_fea = conv3
return output_img, output_fea