-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathnet_lista.py
280 lines (233 loc) · 9.8 KB
/
net_lista.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
# -*- coding:utf-8 -*-
# @Author: Li Hui, Jiangnan University
# @Email: hui_li_jnu@163.com
# @Project : LRRNet
# @File : net_lista.py
# @Time : 2021/4/7 15:26
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
EPSILON = 1e-4
MAX = 1e2
# Convolution operation
class ConvLayer(torch.nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride):
super(ConvLayer, self).__init__()
reflection_padding = int(np.floor(kernel_size / 2))
self.reflection_pad = nn.ReflectionPad2d(reflection_padding)
self.conv2d = nn.Conv2d(in_channels, out_channels, kernel_size, stride)
def forward(self, x):
out = self.reflection_pad(x)
out = self.conv2d(out)
return out
# Dense convolution unit
class DenseConv2d(torch.nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride):
super(DenseConv2d, self).__init__()
self.dense_conv = ConvLayer(in_channels, out_channels, kernel_size, stride)
def forward(self, x):
out = self.dense_conv(x)
out = torch.cat([x, out], 1)
return out
# Dense Block unit
class DenseBlock(torch.nn.Module):
def __init__(self, in_channels, kernel_size, stride):
super(DenseBlock, self).__init__()
out_channels_def = 64
denseblock = []
denseblock += [DenseConv2d(in_channels, out_channels_def, kernel_size, stride),
DenseConv2d(in_channels+out_channels_def, out_channels_def, kernel_size, stride)]
self.denseblock = nn.Sequential(*denseblock)
self.conv = ConvLayer(in_channels+out_channels_def * 2, in_channels, 1, stride)
def forward(self, x, isLast=False):
out = self.denseblock(x)
if not isLast:
out = self.conv(out)
return out
def eta_l1(r_, lam_):
# l_1 norm based
# implement a soft threshold function y=sign(r)*max(0,abs(r)-lam)
B, C, H, W = r_.shape
lam_ = torch.reshape(lam_, [1, C, 1, 1])
lam_ = lam_.repeat(B, 1, H, W)
R = torch.sign(r_) * torch.clamp(torch.abs(r_) - lam_, 0)
return R
class LRR_Block_lista(nn.Module):
def __init__(self, s, n, c, stride):
super(LRR_Block_lista, self).__init__()
self.conv_Wdz = ConvLayer(n, c, s, stride)
self.conv_Wdtz = ConvLayer(c, n, s, stride)
def forward(self, x, tensor_z, lam_theta, lam_z):
# Updating
convZ1 = self.conv_Wdz(tensor_z)
midZ = x - convZ1
tensor_c = lam_z*tensor_z + self.conv_Wdtz(midZ)
# tensor_c = tensor_b + hZ
Z = eta_l1(tensor_c, lam_theta)
return Z
class GetLS_Net(nn.Module):
def __init__(self, s, n, channel, stride, num_block):
super(GetLS_Net, self).__init__()
# n = 128 # number of filters
# s = 3 # filter size
# num_block = 4 # number of layers
# Channel = 3
self.n = n
self.num_block = num_block
self.conv_W00 = ConvLayer(channel, 2*n, s, stride)
self.lamj = nn.Parameter(torch.rand(1, self.n*2)) # l1-norm
self.lamz = nn.Parameter(torch.rand(1, 1))
self.up = nn.Upsample(scale_factor=2)
for i in range(num_block):
self.add_module('lrrblock' + str(i), LRR_Block_lista(s, 2 * n, channel, stride))
def forward(self, x):
b, c, h, w = x.shape
tensor_l = self.conv_W00(x) # Z_0
tensor_z = eta_l1(tensor_l, self.lamj)
for i in range(self.num_block):
# print('num_block - ' + str(i))
lrrblock = getattr(self, 'lrrblock' + str(i))
tensor_z = lrrblock(x, tensor_z, self.lamj, self.lamz)
L = tensor_z[:, :self.n, :, :]
S = tensor_z[:, self.n: 2 * self.n, :, :]
return L, S
class GetLS_dense(nn.Module):
def __init__(self, s, n, channel, stride, num_block):
super(GetLS_dense, self).__init__()
# n = 128 # number of filters
# s = 3 # filter size
# num_block = 4 # number of layers
# Channel = 3
self.num_block = num_block
self.n = n
self.conv_0 = ConvLayer(channel, n, s, stride)
for i in range(num_block):
self.add_module('denseblock' + str(i), DenseBlock(n, s, stride))
def forward(self, x):
tensor_z = self.conv_0(x)
last_block = False
for i in range(self.num_block):
denseblock = getattr(self, 'denseblock' + str(i))
if i == (self.num_block-1):
last_block = True
tensor_z = denseblock(tensor_z, last_block)
L = tensor_z[:, :self.n, :, :]
S = tensor_z[:, self.n: 2 * self.n, :, :]
return L, S
class GetLS_Conv8(nn.Module):
def __init__(self, s, n, channel, stride, num_block):
super(GetLS_Conv8, self).__init__()
# n = 128 # number of filters
# s = 3 # filter size
# num_block = 4 # number of layers
# Channel = 3
self.n = n
self.conv_8 = nn.Sequential(
ConvLayer(channel, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride),
ConvLayer(2 * n, 2 * n, s, stride)
)
def forward(self, x):
tensor_z = self.conv_8(x)
L = tensor_z[:, :self.n, :, :]
S = tensor_z[:, self.n: 2 * self.n, :, :]
return L, S
class Decoder_simple(nn.Module):
def __init__(self, s, n, channel, stride, fusion_type):
super(Decoder_simple, self).__init__()
self.type = fusion_type
self.conv_ReZx = ConvLayer(n, channel, s, stride)
self.conv_ReUx = ConvLayer(n, channel, s, stride)
self.conv_ReZy = ConvLayer(n, channel, s, stride)
self.conv_ReUy = ConvLayer(n, channel, s, stride)
if self.type.__contains__('cat'):
# cat
self.conv_ReL = ConvLayer(2 * channel, channel, s, stride)
self.conv_ReH = ConvLayer(2 * channel, channel, s, stride)
else:
# add
self.conv_ReL = ConvLayer(channel, channel, s, stride)
self.conv_ReH = ConvLayer(channel, channel, s, stride)
def forward(self, Z_x, U_x, Z_y, U_y):
# get loww parts and sparse parts
x_l = self.conv_ReZx(Z_x)
x_h = self.conv_ReUx(U_x)
y_l = self.conv_ReZy(Z_y)
y_h = self.conv_ReUy(U_y)
# reconstructure
if self.type.__contains__('cat'):
# cat
low = self.conv_ReL(torch.cat([x_l, y_l], 1))
high = self.conv_ReH(torch.cat([x_h, y_h], 1))
else:
# add
low = self.conv_ReL(x_l + y_l)
high = self.conv_ReH(x_h + y_h)
out = low + high
return out, x_l, x_h, y_l, y_h, low, high
class LRR_NET(nn.Module):
def __init__(self, s, n, channel, stride, num_block, fusion_type):
super(LRR_NET, self).__init__()
self.get_ls1 = GetLS_Net(s, n, channel, stride, num_block)
self.get_ls2 = GetLS_Net(s, n, channel, stride, num_block)
# self.get_ls1 = GetLS_Conv8(s, n, channel, stride, num_block)
# self.get_ls2 = GetLS_Conv8(s, n, channel, stride, num_block)
#self.get_ls1 = GetLS_dense(s, n, channel, stride, num_block)
#self.get_ls2 = GetLS_dense(s, n, channel, stride, num_block)
self.decoder = Decoder_simple(s, n, channel, stride, fusion_type)
def forward(self, x, y):
fea_x_l, fea_x_s = self.get_ls1(x)
fea_y_l, fea_y_s = self.get_ls2(y)
f, x_l, x_h, y_l, y_h, fl, fh = self.decoder(fea_x_l, fea_x_s, fea_y_l, fea_y_s)
out = {'fea_x_l': fea_x_l, 'fea_x_s': fea_x_s,
'fea_y_l': fea_y_l, 'fea_y_s': fea_y_s,
'x_l': x_l, 'x_h': x_h,
'y_l': y_l, 'y_h': y_h,
'fl': fl, 'fh': fh,
'fuse': f}
return out
class Vgg16(torch.nn.Module):
def __init__(self):
super(Vgg16, self).__init__()
self.conv1_1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1)
self.conv1_2 = nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1)
self.conv2_1 = nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1)
self.conv2_2 = nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1)
self.conv3_1 = nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1)
self.conv3_2 = nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1)
self.conv3_3 = nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1)
self.conv4_1 = nn.Conv2d(256, 512, kernel_size=3, stride=1, padding=1)
self.conv4_2 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1)
self.conv4_3 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1)
self.conv5_1 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1)
self.conv5_2 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1)
self.conv5_3 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1)
def forward(self, X):
b, c, h, w = X.shape
if c == 1:
X = X.repeat(1, 3, 1, 1)
h = F.relu(self.conv1_1(X))
h = F.relu(self.conv1_2(h))
relu1_2 = h
h = F.max_pool2d(h, kernel_size=2, stride=2)
h = F.relu(self.conv2_1(h))
h = F.relu(self.conv2_2(h))
relu2_2 = h
h = F.max_pool2d(h, kernel_size=2, stride=2)
h = F.relu(self.conv3_1(h))
h = F.relu(self.conv3_2(h))
h = F.relu(self.conv3_3(h))
relu3_3 = h
h = F.max_pool2d(h, kernel_size=2, stride=2)
h = F.relu(self.conv4_1(h))
h = F.relu(self.conv4_2(h))
h = F.relu(self.conv4_3(h))
relu4_3 = h
# [relu1_2, relu2_2, relu3_3, relu4_3]
return [relu1_2, relu2_2, relu3_3, relu4_3]