-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathNRKNet.py
182 lines (148 loc) · 6.56 KB
/
NRKNet.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 time
from layers import *
from symcm import *
class EBlock(nn.Module):
def __init__(self, out_channel, num_res=8):
super(EBlock, self).__init__()
layers = [ResBlock(out_channel, out_channel) for _ in range(num_res)]
self.layers = nn.Sequential(*layers)
def forward(self, x):
return self.layers(x)
class DBlock(nn.Module):
def __init__(self, channel, num_res=8):
super(DBlock, self).__init__()
layers = [ResBlock(channel, channel) for _ in range(num_res)]
self.layers = nn.Sequential(*layers)
def forward(self, x):
return self.layers(x)
class NRKNet(nn.Module):
"""
The version of fixed-point iteration with different iteration numbers in different scale
Re-blurred source: the de-blurred image of last scale
Number of scales: 3
Number of iteration in each scale: [1,2,3]
Number of times to estimate the blur kernel in each scale: 1
Normalization form: None
"""
def __init__(self, config):
super(NRKNet, self).__init__()
base_channel = 64
num_res = config.net['num_res']
num_kernels = config.net['num_kernels']
in_ch = config.net['in_ch']
self.SummationLayer = SumLayer(num_kernels + 1)
self.MultiplyLayer = MultiplyLayer()
self.GCM = FullRKR(num_kernels)
self.APU = SqueezeAttentionBlock(base_channel, num_kernels + 1)
self.Encoder = nn.ModuleList([
EBlock(base_channel, num_res),
EBlock(base_channel * 2, num_res),
EBlock(base_channel * 4, num_res),
])
self.Decoder = nn.ModuleList([
DBlock(base_channel * 4, num_res),
DBlock(base_channel * 2, num_res),
DBlock(base_channel, num_res)
])
self.bottleneck = nn.Sequential(
BasicConv(base_channel * 4, base_channel * 4, kernel_size=3, relu=True, stride=1),
EBlock(base_channel * 4, num_res)
)
self.feat_extract = nn.ModuleList([
BasicConv(in_ch, base_channel, kernel_size=3, relu=True, stride=1),
BasicConv(base_channel * 1, base_channel * 2, kernel_size=3, relu=True, stride=2),
BasicConv(base_channel * 2, base_channel * 4, kernel_size=3, relu=True, stride=2),
BasicConv(base_channel * 4, base_channel * 2, kernel_size=4, relu=True, stride=2, transpose=True),
BasicConv(base_channel * 2, base_channel * 1, kernel_size=4, relu=True, stride=2, transpose=True)
])
def forward(self, x, y=None, phase='train', scales=3):
blurrys = []
deblurred = []
items=[]
x_2 = F.interpolate(x, scale_factor=0.5)
x_4 = F.interpolate(x_2, scale_factor=0.5)
xs = [x_4, x_2, x]
if phase == 'train':
y_2 = F.interpolate(y, scale_factor=0.5)
y_4 = F.interpolate(y_2, scale_factor=0.5)
ys = [y_4, y_2, y]
h, c = self.APU.conv_atten.init_hidden(xs[0].shape[0], (xs[0].shape[-2] // 2, xs[0].shape[-1] // 2))
dbd = xs[-1]
for i in range(len(xs)):
'''Feature Extract 0'''
x_ = self.feat_extract[0](xs[i])
res1 = self.Encoder[0](x_)
'''Down Sample 1'''
z = self.feat_extract[1](res1)
res2 = self.Encoder[1](z)
'''Down Sample 2'''
z = self.feat_extract[2](res2)
res3 = self.Encoder[2](z)
'''Bottle Neck'''
z = self.bottleneck(res3)
'''Up Sample 2'''
z = self.Decoder[0](res3 + z)
z = self.feat_extract[3](z)
'''Up Sample 1'''
z = self.Decoder[1](z + res2)
z = self.feat_extract[4](z)
z = self.Decoder[2](z + res1)
z_, h, c = self.APU(z, (h, c))
db_result = xs[i]
temp = xs[i]
for j in range(scales - i):
temp = temp - self.SummationLayer(self.MultiplyLayer(self.GCM(temp), z_))
db_result = db_result + temp
items.append(temp)
dbd = dbd + F.interpolate(temp, scale_factor=2 ** (scales - i - 1), mode='bilinear')
deblurred.append(db_result)
blur = self.SummationLayer(self.MultiplyLayer(self.GCM(ys[i]), z_))
blurrys.append(blur)
h = F.interpolate(h, scale_factor=2, mode='bilinear')
c = F.interpolate(c, scale_factor=2, mode='bilinear')
deblurred.append(dbd)
return deblurred, blurrys
else:
h, c = self.APU.conv_atten.init_hidden(xs[0].shape[0], (xs[0].shape[-2] // 2, xs[0].shape[-1] // 2))
dbd = xs[-1]
for i in range(len(xs)):
'''Feature Extract 0'''
x_ = self.feat_extract[0](xs[i])
res1 = self.Encoder[0](x_)
'''Down Sample 1'''
z = self.feat_extract[1](res1)
res2 = self.Encoder[1](z)
'''Down Sample 2'''
z = self.feat_extract[2](res2)
res3 = self.Encoder[2](z)
'''Bottle Neck'''
z = self.bottleneck(res3)
'''Up Sample 2'''
z = self.Decoder[0](res3 + z)
z = self.feat_extract[3](z)
'''Up Sample 1'''
z = self.Decoder[1](z + res2)
z = self.feat_extract[4](z)
z = self.Decoder[2](z + res1)
z_, h, c = self.APU(z, (h, c))
temp = xs[i]
for j in range(scales - i):
temp = temp - self.SummationLayer(self.MultiplyLayer(self.GCM(temp), z_))
items.append(temp)
dbd = dbd + F.interpolate(temp, scale_factor=2 ** (scales - i - 1), mode='bilinear')
if i < scales - 1:
h = F.interpolate(h, scale_factor=2, mode='bilinear')
c = F.interpolate(c, scale_factor=2, mode='bilinear')
deblurred.append(
xs[-1] + items[-1] + F.interpolate(items[-2], scale_factor=2, mode='bilinear') + F.interpolate(
items[-3],
scale_factor=4,
mode='bilinear'))
return deblurred, blurrys
if __name__ == '__main__':
import config as config
from thop import profile
kmlnet = NRKNet(config).cuda()
a = torch.rand((1, 3, 1280, 720)).cuda()
flops, params = profile(kmlnet, inputs=(a,None,'test'))
print(flops/2, params)