-
Notifications
You must be signed in to change notification settings - Fork 8
/
model_Dense_w_net_modelfy_dialate.py
executable file
·305 lines (226 loc) · 10.7 KB
/
model_Dense_w_net_modelfy_dialate.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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
from networks import *
from Utils.utils import *
import torch
from torch.nn.functional import interpolate as Inter
class Rain_steaks_with_BG(nn.Module):
'''get rain steaks with background from RGBrain_img
return img-----512 512 3*3 '''
def __init__(self):
super(Rain_steaks_with_BG, self).__init__()
self.conv_3_0 = conv_blocks_size_3(in_dim=3 , out_dim=4 , Use_pool=True) #256
weights_init_xavier(self.conv_3_0)
self.conv_3_1 = conv_blocks_size_3(4,8,Use_pool=True) #128
weights_init_xavier(self.conv_3_1)
self.conv_3_2 = conv_blocks_size_3(8,16,True) #64
weights_init_xavier(self.conv_3_2)
self.deconv_3_0 = deconv_blocks_size_3(16,8,True) #128
weights_init_xavier(self.deconv_3_0)
self.deconv_3_1 = deconv_blocks_size_3(8, 4, True) #256
weights_init_xavier(self.deconv_3_1)
self.deconv_3_2 = deconv_blocks_size_3(4,3,True) #512
weights_init_xavier(self.deconv_3_2)
self.conv_7_0 = conv_blocks_size3_dilation2(in_dim=3, out_dim=4, Use_pool=True) # 256
weights_init_xavier(self.conv_7_0)
self.conv_7_1 = conv_blocks_size3_dilation2(4, 8, Use_pool=True) # 128
weights_init_xavier(self.conv_7_1)
self.conv_7_2 = conv_blocks_size3_dilation2(8,16,True)
weights_init_xavier(self.conv_7_2)
# self.conv_5_3 = conv_blocks_size_5(16, 32, False) # 64
# weights_init_xavier(self.conv_5_3)
self.conv_9_0 = conv_blocks_size3_dilation3(in_dim=3, out_dim=4, Use_pool=True) # 256
weights_init_xavier(self.conv_9_0)
self.conv_9_1 = conv_blocks_size3_dilation3(4,8, Use_pool=True) # 128
weights_init_xavier(self.conv_9_1)
self.conv_9_2 = conv_blocks_size3_dilation3(8,16, True) # 64
weights_init_xavier(self.conv_9_2)
# self.conv_7_3 = conv_blocks_size_7(16, 32, False) # 32
# weights_init_xavier(self.conv_7_3)
def forward(self,x):
input_3 = x
input_3 = self.conv_3_0(input_3)
input_3 = self.conv_3_1(input_3)
input_3 = self.conv_3_2(input_3)
input_3 = self.deconv_3_0(input_3)
input_3 = self.deconv_3_1(input_3)
input_3 = self.deconv_3_2(input_3)
input_5 = x
input_5 = self.conv_7_0(input_5)
# input_5 = self.Dropout(input_5)
input_5 = self.conv_7_1(input_5)
# input_5 = self.Dropout(input_5)
input_5 = self.conv_7_2(input_5)
input_5 = self.deconv_3_0(input_5)
input_5 = self.deconv_3_1(input_5)
input_5 = self.deconv_3_2(input_5)
input_7 = x
input_7 = self.conv_9_0(input_7)
input_7 = self.conv_9_1(input_7)
input_7 = self.conv_9_2(input_7)
input_7 = self.deconv_3_0(input_7)
input_7 = self.deconv_3_1(input_7)
input_7 = self.deconv_3_2(input_7)
return torch.cat([input_3,input_5,input_7] ,dim=1)
class Low_BackGround(nn.Module):
'''get low background from RGB rain_Img
in order to move rain steaks ,wo decide to insert some residual img
to make CNN do his jod better input 512 512 3 retrun 32 32 64'''
def __init__(self):
super(Low_BackGround, self).__init__()
self.encode1 = conv_blocks_size_3(in_dim=4 , out_dim=8 ,Use_pool=True) #256
weights_init_xavier(self.encode1)
self.encode2 = conv_blocks_size_3(in_dim=8 ,out_dim=16,Use_pool=True) #128
weights_init_xavier(self.encode2)
self.encode3 = conv_blocks_size_3(in_dim=16 ,out_dim=32,Use_pool=True) #64
weights_init_xavier(self.encode3)
self.encode4 = conv_blocks_size_3(in_dim=32,out_dim=64,Use_pool=True) #32
weights_init_xavier(self.encode4)
self.encode5 = conv_blocks_size_3(in_dim=64,out_dim=128,Use_pool=True) #16
weights_init_xavier(self.encode5)
# self.encode6 = conv_blocks_size_5(in_dim=128,out_dim=256,Use_pool=True)#8
# weights_init_xavier(self.encode5)
# self.decode1 = deconv_blocks_size_3(in_dim=256,out_dim=128,Use_pool=True) #16
# weights_init_xavier(self.decode1)
self.decode2 =deconv_blocks_size_3(128,64,Use_pool=True) #32
weights_init_xavier(self.decode2)
self.decode3 = deconv_blocks_size_3(64,32,Use_pool=True) #64
weights_init_xavier(self.decode3)
self.decode4 = deconv_blocks_size_3(32,16,Use_pool=True) #128
weights_init_xavier(self.decode4)
self.decode5 = deconv_blocks_size_3(16,8,True) #256
weights_init_xavier(self.decode5)
self.decode6 = deconv_blocks_size_3(8,4,True) #512
weights_init_xavier(self.decode6)
self.eencode1 = conv_blocks_size_3(4,8,True) #256
weights_init_xavier(self.eencode1)
self.eencode2 = conv_blocks_size_3(8,16,True)#128
weights_init_xavier(self.eencode2)
self.eencode3 = conv_blocks_size_3(16,32,True)#64
weights_init_xavier(self.eencode3)
self.eencode4 = conv_blocks_size_3(32, 64, True) # 32
weights_init_xavier(self.eencode4)
self.eencode5 = conv_blocks_size_3(64, 128, True) # 16
weights_init_xavier(self.eencode5)
self.ddecode1 = deconv_blocks_size_3(128, 64, Use_pool=True) # 32
weights_init_xavier(self.ddecode1)
self.ddecode2 = deconv_blocks_size_3(64, 32, Use_pool=True) # 64
weights_init_xavier(self.ddecode2)
self.ddecode3 = deconv_blocks_size_3(32, 16, Use_pool=True) # 128
weights_init_xavier(self.ddecode3)
self.ddecode4 = deconv_blocks_size_3(16, 8, Use_pool=True) # 256
weights_init_xavier(self.ddecode4)
self.ddecode5 = deconv_blocks_size_3(8, 4, Use_pool=True) # 512
weights_init_xavier(self.ddecode5)
def forward(self,x):
# x 512 512 3
"""insert img to use residual_img as a auxiliary"""
insert_img = 0.8 * get_residue(x) + 0.2 * rgb2grad(x) #insert image
x = torch.cat([x , insert_img], 1) #512
del insert_img
encode1 = self.encode1(x) #256 256 8
encode2 = self.encode2(encode1) #128 128 16
encode3 = self.encode3(encode2) #64 64 32
encode4 = self.encode4(encode3) #32 32 64
encode5 = self.encode5(encode4) #16 16 128
# encode6 = self.encode6(encode5) #8
# decode1 = torch.add(encode5 , self.decode1(encode6)) #16
decode = torch.add(encode4,self.decode2(encode5)) # 32
decode = torch.add(encode3,self.decode3(decode)) # 64
decode = torch.add(encode2,self.decode4(decode)) # 128
decode = torch.add(encode1,self.decode5(decode)) # 256
decode = self.decode6(decode)
eencode = torch.add(self.eencode1(decode), encode1) # 256
eencode = torch.add(self.eencode2(eencode) , encode2) #128
eencode = torch.add(self.eencode3(eencode) , encode3) #64
eencode = torch.add(self.eencode4(eencode) , encode4) #32
eencode = torch.add(self.eencode5(eencode) , encode5) #16
del encode5 , decode
ddecode = torch.add(encode4 , self.ddecode1(eencode)) #32
ddecode = torch.add(encode3, self.ddecode2(ddecode)) # 64
ddecode = torch.add(encode2, self.ddecode3(ddecode)) # 128
ddecode = torch.add(encode1, self.ddecode4(ddecode)) # 256
ddecode = self.ddecode5(ddecode) # 512
return ddecode
class Classfication(nn.Module):
'''classfication to classfy rain dense ,make sure Rain_with_Background success extract
rain steaks (with some high level background will be fine )
input 512 512 9'''
def __init__(self):
super(Classfication, self).__init__()
self.features = nn.Sequential(
nn.Conv2d(9,16, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(16),
nn.MaxPool2d(kernel_size=2, stride=2), #256
nn.Conv2d(16,32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(32),
nn.MaxPool2d(kernel_size=2, stride=2), #128
nn.Conv2d(32, 2*32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(2*32),
nn.MaxPool2d(kernel_size=2, stride=2), # 64
nn.Conv2d(32*2, 4 * 32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d( 4 * 32),
nn.MaxPool2d(kernel_size=2, stride=2), # 32
nn.Conv2d(32 * 4, 8 * 32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(8 * 32),
nn.MaxPool2d(kernel_size=2, stride=2), # 16
nn.Conv2d(32 * 8, 16 * 32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(16 * 32),
nn.MaxPool2d(kernel_size=2, stride=2), # 8
nn.Conv2d(32 * 16, 32 * 32, kernel_size=3, stride=1, padding=1),
nn.ReLU(True),
nn.BatchNorm2d(32 * 32),
nn.MaxPool2d(kernel_size=2, stride=2), # 4
)
self.classfier = nn.Sequential(
nn.Linear( 4*4*32*32, 4096 ),
nn.ReLU(True),
nn.Linear(4096 , 1000 ),
nn.ReLU(True),
nn.Linear(1000 ,3),
nn.ReLU(True),
)
weights_init_xavier(self.features)
weights_init_xavier(self.classfier)
self.relu = nn.ReLU()
def forward(self,x):
x = self.relu(x)
x = self.features(x)
x = x.view( x.size(0) , -1)
x = self.classfier(x)
return x
class Refine(nn.Module):
'''refine our model out put ,ny magnifying input and then refine it
input 64 64 4*32 return 512 512 3 '''
def __init__(self):
super(Refine, self).__init__()
self.high1 = conv_blocks_size_3(9,3,bn = False)
weights_init_xavier(self.high1)
# self.low2 = deconv_blocks_size_3(16 ,8 ,True)
# weights_init_xavier(self.low2)
# self.low3 = deconv_blocks_size_3(8,4,True)
# weights_init_xavier(self.low3)
self.conv1 =conv_blocks_size_3(10,5,bn = False)
weights_init_xavier(self.conv1)
self.conv2 = conv_blocks_size_3(5, 3,bn = False)
weights_init_xavier(self.conv2)
# self.conv3 = conv_blocks_size_3(4, 3,bn = False)
# weights_init_xavier(self.conv3)
self.relu = Nonlinear_layer(relu= True)
def forward(self,High, Low , rain_img):
'''High 512 512 9
Low 512 512 4 '''
#
# Low = self.low1(Low) # 128
# Low = self.low2(Low) # 256
# Low = self.low3(Low) #512
High = self.high1(High)
rain_img = torch.cat([Low , High , rain_img],dim=1) #512 512 10
rain_img = self.conv1(rain_img)
rain_img = self.conv2(rain_img)
# rain_img = self.conv3(rain_img)
return rain_img