-
Notifications
You must be signed in to change notification settings - Fork 10
/
architecture.py
171 lines (107 loc) · 4.37 KB
/
architecture.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
class ConvBlock(nn.Module):
def __init__(self, c_in, c_out, ks, k_stride=1):
super().__init__()
self.conv1 = nn.Conv3d(c_in, c_out, ks, stride=k_stride, padding=(1, 1, 1))
self.bn = nn.BatchNorm3d(c_out)
self.elu = nn.ELU()
self.pool = nn.MaxPool3d(kernel_size=(3, 3, 3), stride=2)
self.dropout = nn.Dropout3d(p=0.1)
def forward(self, x):
out = self.conv1(x)
out = self.bn(out)
out = self.elu(out)
out = self.pool(out)
out = self.dropout(out)
return out
class ConvBlock2(nn.Module):
def __init__(self, chans, ks, k_stride=1):
super().__init__()
conv1_ks = [1] + ks[:-1]
conv2_ks = [ks[-1]] + [1,1]
#depthwise
self.conv1 = nn.Conv3d(chans, chans, kernel_size=conv1_ks, stride=k_stride, padding=(0,1,1), groups=chans)
#pointwise
self.conv2 = nn.Conv3d(chans, chans, kernel_size=conv2_ks, stride=k_stride, padding=(1,0,0))
self.bn = nn.BatchNorm3d(chans)
self.elu = nn.ELU()
self.dropout = nn.Dropout3d(p=0.1)
def forward(self, x):
out = self.conv1(x)
out = self.conv2(out)
out = self.bn(out)
out = self.elu(out)
out = self.dropout(out)
return out
class FCBlock(nn.Module):
def __init__(self, chan_in, units_out):
super().__init__()
self.fc = nn.Linear(chan_in, units_out)
self.bn = nn.BatchNorm1d(units_out)
self.elu = nn.ELU()
self.dropout = nn.Dropout(p=0.1)
def forward(self, x):
out = self.fc(x)
out = self.bn(out)
out = self.elu(out)
out = self.dropout(out)
return out
class Camull(nn.Module):
def __init__(self):
super().__init__()
self.stack1 = nn.Sequential(ConvBlock(1, 24, (11,11,11), 2),
ConvBlock(24, 48, (5,5,5)))
self.stack1_b = nn.Sequential(ConvBlock(1, 24, (11,11,11), 2),
ConvBlock(24, 48, (5,5,5)))
#seperable convolutions
self.stack2 = nn.Sequential(ConvBlock2(96, [3,3,3]),
ConvBlock2(96, [3,3,3]),
ConvBlock2(96, [3,3,3]))
#Number of channels inputted is halfed to reduce number of parameters.
#This is done to the input in the forward function.
self.stack3_a = nn.Sequential(ConvBlock(48, 24, (3,3,3)),
ConvBlock(24, 8, (3,3,3)))
self.stack3_b = nn.Sequential(ConvBlock(48, 24, (3,3,3)),
ConvBlock(24, 8, (3,3,3)))
self.fcblock = nn.Sequential(FCBlock(21, 32),
FCBlock(32, 10))
self.flat = nn.Flatten()
self.fc1 = FCBlock(128, 10)
self.fc2 = FCBlock(20, 4)
self.lin = nn.Linear(4, 1)
self.sig = nn.Sigmoid()
#Performing a grouped convolutional stack
def s3_forward(self, x):
bound = int(np.floor(x.shape[1]/2))
out_a = x[:,:bound]
out_b = x[:,bound:]
out_a = self.stack3_a(out_a)
out_b = self.stack3_b(out_b)
out = torch.cat((out_a, out_b), 1)
return out
def cat_with_clin(self, x_a, x_b):
out = self.flat(x_a)
out_a = self.fc1(out)
out_b = self.fcblock(x_b)
out = torch.cat((out_a, out_b), 1)
return out
def forward(self, x):
mri, clin = x
out_a = self.stack1(mri)
out_b = self.stack1_b(mri)
out = torch.cat((out_a, out_b), 1) #1 as ind 0 is batch size
identity = out
out = self.stack2(out)
out = out + identity
out = self.s3_forward(out)
out = self.cat_with_clin(out, clin)
out = self.fc2(out)
out = self.lin(out)
out = self.sig(out)
return out
def load_cam_model(path):
model = torch.load(path)
return model