-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathResUNet_3D.py
147 lines (127 loc) · 6.29 KB
/
ResUNet_3D.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
# Author： Dongdong Zhao
# 参照Kits19比赛中 top 1 方案写的网络模型
import torch
import torch.nn as nn
from model_zoo.BaseModelClass import BaseModel
class ResidualBlock_down(nn.Module):
def __init__(self, in_channels, out_channels):
super(ResidualBlock_down, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.down = nn.Conv3d(self.in_channels, self.out_channels, 2, 2)
self.res_block = ResidualBlock(self.out_channels, self.out_channels)
def forward(self, x):
out = self.down(x)
out = self.res_block(out)
return out
class ResidualBlock_up(nn.Module):
def __init__(self, in_channels, out_channels):
super(ResidualBlock_up, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.up = nn.ConvTranspose3d(self.in_channels,self.out_channels, 2, 2)
self.res_block = ResidualBlock(self.out_channels, self.out_channels)
def forward(self, x):
out = self.up(x)
out = self.res_block(out)
return out
# class ResidualBlock(nn.Module):
# def __init__(self, in_channels, out_channels):
# super(ResidualBlock, self).__init__()
# self.in_channels = in_channels
# self.out_channels = out_channels
# self.res_conv = nn.Sequential(nn.Conv3d(self.in_channels, self.out_channels, 3, 1, padding=1),
# nn.InstanceNorm3d(self.out_channels),
# nn.ReLU(inplace=True),
# nn.Conv3d(self.out_channels, self.out_channels, 3, 1, padding=1),
# nn.InstanceNorm3d(self.out_channels)
# )
# self.conv_extral = nn.Conv3d(self.out_channels, self.in_channels, 3, 1, padding=1)
#
# def forward(self, x):
# out = self.res_conv(x)
# if x.shape != out.shape:
# out = self.conv_extral(out)
# out = out + x
# return out
class ResidualBlock(nn.Module):
def __init__(self, in_channels, out_channels):
super(ResidualBlock, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.res_conv = nn.Sequential(nn.Conv3d(self.in_channels, self.in_channels // 2, 1, 1),
nn.InstanceNorm3d(self.out_channels),
nn.ReLU(inplace=True),
nn.Conv3d(self.in_channels // 2, self.in_channels // 2, 3, 1, padding=1),
nn.InstanceNorm3d(self.out_channels),
nn.ReLU(inplace=True),
nn.Conv3d(self.in_channels // 2, self.out_channels, 1, 1),
nn.InstanceNorm3d(self.out_channels)
)
self.conv_extral = nn.Conv3d(self.out_channels, self.in_channels, 3, 1, padding=1)
self.relu = nn.ReLU(in_channels)
def forward(self, x):
out = self.res_conv(x)
if x.shape != out.shape:
out = self.conv_extral(out)
out = out + x
out = self.relu(out)
return out
class ResUNet_3D(BaseModel):
def __init__(self, in_channels=1, out_channels=1):
super(ResUNet_3D, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.in_conv = nn.Sequential(nn.Conv3d(self.in_channels, 30, 3, 1, padding=1),
nn.InstanceNorm3d(30),
nn.ReLU(inplace=True))
self.residual_block1 = ResidualBlock(30, 30)
self.down1 = ResidualBlock_down(30, 60)
self.residual_block2 = ResidualBlock(60, 60)
self.down2 = ResidualBlock_down(60, 120)
self.residual_block3 = ResidualBlock(120, 120)
self.down3 = ResidualBlock_down(120, 240)
self.residual_block4 = ResidualBlock(240, 240)
self.down4 = ResidualBlock_down(240, 480)
self.residual_block5 = ResidualBlock(480, 480)
self.up1 = ResidualBlock_up(480, 240)
self.up2 = ResidualBlock_up(480, 120)
self.up3 = ResidualBlock_up(240, 60)
self.up4 = ResidualBlock_up(120, 30)
self.out_conv = nn.Conv3d(60, 1, 1)
def forward(self, x):
out1 = self.in_conv(x) # 对输入进行卷积—>（1， 30， 64， 64， 64）
out1 = self.residual_block1(out1)
out2 = self.down1(out1) # 第一次下采样完成 # （1， 60， 32， 32，32）
out2 = self.residual_block2(out2)
out3 = self.down2(out2) # 第二次下采样完成 # （1， 120， 16， 16，16）
out3 = self.residual_block3(out3)
out4 = self.down3(out3) # 第三次下采样完成 # （1， 240， 8， 8，8）
out4 = self.residual_block4(out4)
out5 = self.down4(out4) # （1， 480， 4， 4，4）
core_feature = self.residual_block5(out5) # （1， 480， 4， 4，4）
out5 = self.up1(core_feature) # (1, 240, 8, 8, 8)
out5 = torch.cat((out5, out4), dim=1) # # (1, 480, 8, 8, 8)
out6 = self.up2(out5) # (1, 120, 16, 16, 16)
out6 = torch.cat((out6, out3), dim=1) #(1, 240, 16, 16, 16)
out7 = self.up3(out6) # (1, 60, 32, 32, 32)
out7 = torch.cat((out7, out2), dim=1) # (1, 120, 32, 32, 32)
out8 = self.up4(out7) #(1, 30, 64, 64, 64)
out8 = torch.cat((out8, out1), dim=1) #(1, 60, 64, 64, 64)
out = self.out_conv(out8)
out = nn.Sigmoid()(out)
return out
def test(self,device='cpu'):
input_tensor = torch.rand(1, 2, 32, 32, 32)
ideal_out = torch.rand(1, self.n_classes, 32, 32, 32)
out = self.forward(input_tensor)
assert ideal_out.shape == out.shape
# summary(self.to(torch.device(device)), (2, 32, 32, 32),device='cpu')
# import torchsummaryX
# torchsummaryX.summary(self, input_tensor.to(device))
print("ResUNet_2stage test is complete")
if __name__ == '__main__':
input = torch.randn(1, 1, 64, 64, 64)
net = ResUNet_3D()
out = net(input)
print(out.shape)