-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathnew_models.py
201 lines (185 loc) · 9.3 KB
/
new_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
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
# This is the file where the actual architecture of the network is written
# This file uses the module classes which are defined in the seg_modules.py file and builds upon that
# Links to each of the architecture are given in the train_parameters.cfg file
import torch.nn.functional as F
import torch.nn as nn
import torch
from seg_modules import in_conv, DownsamplingModule, EncodingModule, InceptionModule, ResNetModule
from seg_modules import UpsamplingModule, DecodingModule,IncDownsamplingModule,IncConv
from seg_modules import out_conv, FCNUpsamplingModule, IncDropout,IncUpsamplingModule
class unet(nn.Module):
def __init__(self, n_channels, n_classes, base_filters):
super(unet, self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.ins = in_conv(self.n_channels, base_filters)
self.ds_0 = DownsamplingModule(base_filters, base_filters*2)
self.en_1 = EncodingModule(base_filters*2, base_filters*2)
self.ds_1 = DownsamplingModule(base_filters*2, base_filters*4)
self.en_2 = EncodingModule(base_filters*4, base_filters*4)
self.ds_2 = DownsamplingModule(base_filters*4, base_filters*8)
self.en_3 = EncodingModule(base_filters*8, base_filters*8)
self.ds_3 = DownsamplingModule(base_filters*8, base_filters*16)
self.en_4 = EncodingModule(base_filters*16, base_filters*16)
self.us_3 = UpsamplingModule(base_filters*16, base_filters*8)
self.de_3 = DecodingModule(base_filters*16, base_filters*8)
self.us_2 = UpsamplingModule(base_filters*8, base_filters*4)
self.de_2 = DecodingModule(base_filters*8, base_filters*4)
self.us_1 = UpsamplingModule(base_filters*4, base_filters*2)
self.de_1 = DecodingModule(base_filters*4, base_filters*2)
self.us_0 = UpsamplingModule(base_filters*2, base_filters)
self.out = out_conv(base_filters*2, self.n_classes)
def forward(self, x):
x1 = self.ins(x)
x2 = self.ds_0(x1)
x2 = self.en_1(x2)
x3 = self.ds_1(x2)
x3 = self.en_2(x3)
x4 = self.ds_2(x3)
x4 = self.en_3(x4)
x5 = self.ds_3(x4)
x5 = self.en_4(x5)
x = self.us_3(x5)
x = self.de_3(x, x4)
x = self.us_2(x)
x = self.de_2(x, x3)
x = self.us_1(x)
x = self.de_1(x, x2)
x = self.us_0(x)
x = self.out(x, x1)
return x
class resunet(nn.Module):
def __init__(self, n_channels, n_classes, base_filters):
super(resunet, self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.ins = in_conv(self.n_channels, base_filters, res=True)
self.ds_0 = DownsamplingModule(base_filters, base_filters*2)
self.en_1 = EncodingModule(base_filters*2, base_filters*2, res=True)
self.ds_1 = DownsamplingModule(base_filters*2, base_filters*4)
self.en_2 = EncodingModule(base_filters*4, base_filters*4, res=True)
self.ds_2 = DownsamplingModule(base_filters*4, base_filters*8)
self.en_3 = EncodingModule(base_filters*8, base_filters*8, res=True)
self.ds_3 = DownsamplingModule(base_filters*8, base_filters*16)
self.en_4 = EncodingModule(base_filters*16, base_filters*16, res=True)
self.us_3 = UpsamplingModule(base_filters*16, base_filters*8)
self.de_3 = DecodingModule(base_filters*16, base_filters*8, res=True)
self.us_2 = UpsamplingModule(base_filters*8, base_filters*4)
self.de_2 = DecodingModule(base_filters*8, base_filters*4, res=True)
self.us_1 = UpsamplingModule(base_filters*4, base_filters*2)
self.de_1 = DecodingModule(base_filters*4, base_filters*2, res=True)
self.us_0 = UpsamplingModule(base_filters*2, base_filters)
self.out = out_conv(base_filters*2, self.n_classes, res=True)
def forward(self, x):
x1 = self.ins(x)
x2 = self.ds_0(x1)
x2 = self.en_1(x2)
x3 = self.ds_1(x2)
x3 = self.en_2(x3)
x4 = self.ds_2(x3)
x4 = self.en_3(x4)
x5 = self.ds_3(x4)
x5 = self.en_4(x5)
x = self.us_3(x5)
x = self.de_3(x, x4)
x = self.us_2(x)
x = self.de_2(x, x3)
x = self.us_1(x)
x = self.de_1(x, x2)
x = self.us_0(x)
x = self.out(x, x1)
return x
class fcn(nn.Module):
def __init__(self, n_channels, n_classes, base_filters):
super(fcn, self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.ins = in_conv(self.n_channels, base_filters)
self.ds_0 = DownsamplingModule(base_filters, base_filters*2)
self.en_1 = EncodingModule(base_filters*2, base_filters*2)
self.ds_1 = DownsamplingModule(base_filters*2, base_filters*4)
self.en_2 = EncodingModule(base_filters*4, base_filters*4)
self.ds_2 = DownsamplingModule(base_filters*4, base_filters*8)
self.en_3 = EncodingModule(base_filters*8, base_filters*8)
self.ds_3 = DownsamplingModule(base_filters*8, base_filters*16)
self.en_4 = EncodingModule(base_filters*16, base_filters*16)
self.us_4 = FCNUpsamplingModule(base_filters*16, 1, scale_factor=5)
self.us_3 = FCNUpsamplingModule(base_filters*8, 1, scale_factor=4)
self.us_2 = FCNUpsamplingModule(base_filters*4, 1, scale_factor=3)
self.us_1 = FCNUpsamplingModule(base_filters*2, 1, scale_factor=2)
self.us_0 = FCNUpsamplingModule(base_filters, 1, scale_factor=1)
self.conv_0 = nn.Conv3d(in_channels=5, out_channels=self.n_classes,
kernel_size=1, stride=1, padding=0, bias=True)
def forward(self, x):
x1 = self.ins(x)
x2 = self.ds_0(x1)
x2 = self.en_1(x2)
x3 = self.ds_1(x2)
x3 = self.en_2(x3)
x4 = self.ds_2(x3)
x4 = self.en_3(x4)
x5 = self.ds_3(x4)
x5 = self.en_4(x5)
u5 = self.us_4(x5)
u4 = self.us_3(x4)
u3 = self.us_2(x3)
u2 = self.us_1(x2)
u1 = self.us_0(x1)
x = torch.cat([u5, u4, u3, u2, u1], dim=1)
x = self.conv_0(x)
return F.softmax(x,dim=1)
#defining the inception net class which inherits from the nn.module class - which is I think to basically tell the program that this is a pytorch network model that I am defining
#the res parameter is for the addition of the initial feature map with the final feature map after performance of the convolutions - for the decoding module, not the initial input but the input after the first convolution is addded to the final output since the initial input and the final one do not have the same dimensions
#In the init function we just define the operations that are going to be performed in the forward pass of the network - and in the forward method we actually substitute the values in the defined operations
class uinc(nn.Module):
def __init__(self,n_channels,n_classes,base_filters):
super(uinc,self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.conv0_1x1 = IncConv(n_channels, base_filters)
self.rn_0 = ResNetModule(base_filters,base_filters,res=True)
self.ri_0 = InceptionModule(base_filters,base_filters,res=True)
self.ds_0 = IncDownsamplingModule(base_filters,base_filters*2)
self.ri_1 = InceptionModule(base_filters*2,base_filters*2,res=True)
self.ds_1 = IncDownsamplingModule(base_filters*2,base_filters*4)
self.ri_2 = InceptionModule(base_filters*4,base_filters*4,res=True)
self.ds_2 = IncDownsamplingModule(base_filters*4,base_filters*8)
self.ri_3 = InceptionModule(base_filters*8,base_filters*8,res=True)
self.ds_3 = IncDownsamplingModule(base_filters*8,base_filters*16)
self.ri_4 = InceptionModule(base_filters*16,base_filters*16,res=True)
self.us_3 = IncUpsamplingModule(base_filters*16,base_filters*8)
self.ri_5 = InceptionModule(base_filters*16,base_filters*16,res=True)
self.us_2 = IncUpsamplingModule(base_filters*16,base_filters*4)
self.ri_6 = InceptionModule(base_filters*8,base_filters*8,res=True)
self.us_1 = IncUpsamplingModule(base_filters*8,base_filters*2)
self.ri_7 = InceptionModule(base_filters*4,base_filters*4,res=True)
self.us_0 = IncUpsamplingModule(base_filters*4,base_filters)
self.ri_8 = InceptionModule(base_filters*2,base_filters*2,res=True)
self.conv9_1x1 = IncConv(base_filters*2,base_filters)
self.rn_10 = ResNetModule(base_filters*2,base_filters*2,res=True)
self.dropout = IncDropout(base_filters*2,n_classes)
def forward(self,x):
x = self.conv0_1x1(x)
x1 = self.rn_0(x)
x2 = self.ri_0(x1)
x3 = self.ds_0(x2)
x3 = self.ri_1(x3)
x4 = self.ds_1(x3)
x4 = self.ri_2(x4)
x5 = self.ds_2(x4)
x5 = self.ri_3(x5)
x6 = self.ds_3(x5)
x6 = self.ri_4(x6)
x6 = self.us_3(x6)
x6 = self.ri_5(torch.cat((x5,x6),dim=1))
x6 = self.us_2(x6)
x6 = self.ri_6(torch.cat((x4,x6),dim=1))
x6 = self.us_1(x6)
x6 = self.ri_7(torch.cat((x3,x6),dim=1))
x6 = self.us_0(x6)
x6 = self.ri_8(torch.cat((x2,x6),dim=1))
x6 = self.conv9_1x1(x6)
x6 = self.rn_10(torch.cat((x1,x6),dim=1))
x6 = self.dropout(x6)
x6 = F.softmax(x6,dim=1)
return x6