-
Notifications
You must be signed in to change notification settings - Fork 0
/
Unet_model.py
144 lines (120 loc) · 4.69 KB
/
Unet_model.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
# -*- coding: utf-8 -*-
"""
@author: qing
usage:define U-Net model and test
"""
import torch
import torch.utils.data
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import numpy as np
import time
import os
class Unet(nn.Module):
def convlayer(self,in_channel,out_channel):
#两侧中的同级内卷积运算模块
#in_channel:输入图片通道数;out_channel:输出图片通道数
module=nn.Sequential(
#conv:o=(i-k+2*p)/s+1
nn.Conv2d(in_channel,out_channel,kernel_size=3,stride=1,padding=1),
nn.BatchNorm2d(out_channel),
nn.ReLU(inplace=True),
nn.Conv2d(out_channel,out_channel,kernel_size=3,stride=1,padding=1),
nn.BatchNorm2d(out_channel),
nn.ReLU(inplace=True)
)
return module
def maxpool(self,x,kernel_size,stride,padding):
#左半侧编码器的(下采样)最大池化操作
return nn.MaxPool2d(kernel_size,stride,padding)(x)
def deconvlayer(self,in_channel,out_channel,k,s,p):
#右半侧解码器的上采样(反卷积)操作
module=nn.Sequential(
#deconv:o=(i-1)s+k-2*p
nn.ConvTranspose2d(in_channel,out_channel,kernel_size=k,stride=s,padding=p),
nn.ReLU(inplace=True)
)
return module
def __init__(self):
super(Unet,self).__init__()
#左半侧编码器结构的卷积运算
self.D_conv1=self.convlayer(3,8)
self.D_conv2=self.convlayer(8,16)
self.D_conv3=self.convlayer(16,32)
self.D_conv4=self.convlayer(32,64)
self.B_conv=self.convlayer(64,128)
#右半侧解码器结构的卷积运算
self.U_conv4=self.convlayer(128,64)
self.U_conv3=self.convlayer(64,32)
self.U_conv2=self.convlayer(32,16)
self.U_conv1=self.convlayer(16,8)
self.last=nn.Conv2d(8,2,kernel_size=1,stride=1)
#右半侧解码器结构的上采样运算
self.U4=self.deconvlayer(128,64,2,2,1)
self.U3=self.deconvlayer(64,32,2,2,1)
self.U2=self.deconvlayer(32,16,2,2,1)
self.U1=self.deconvlayer(16,8,2,2,0)
def forward(self,x):
#编码器部分
d1=self.D_conv1(x)
#print('d1 shape:{}'.format(d1.size()))
d1_pool=self.maxpool(d1,2,2,0)
#print('d1_pool shape:{}'.format(d1_pool.size()))
d2=self.D_conv2(d1_pool)
#print('d2 shape:{}'.format(d2.size()))
d2_pool=self.maxpool(d2,2,2,1)
#print('d2_pool shape:{}'.format(d2_pool.size()))
d3=self.D_conv3(d2_pool)
#print('d3 shape:{}'.format(d3.size()))
d3_pool=self.maxpool(d3,2,2,1)
#print('d3_pool shape:{}'.format(d3_pool.size()))
d4=self.D_conv4(d3_pool)
#print('d4 shape:{}'.format(d4.size()))
d4_pool=self.maxpool(d4,2,2,1)
#print('d4_pool shape:{}'.format(d4_pool.size()))
d5=self.B_conv(d4_pool)
#print('d5 shape:{}'.format(d5.size()))
#解码器部分
u4=self.U4(d5)
u4=u4.contiguous()
#print('u4 shape:{}'.format(u4.size()))
#两个不同的张量按高度维度堆叠
u4_=self.U_conv4(torch.cat((d4,u4),1))
#print('u4_ shape:{}'.format(u4_.size()))
#print('u4_ size:{}'.format(u4_.size()))
u3=self.U3(u4_)
u3=u3.contiguous()
#print('u3 shape:{}'.format(u3.size()))
u3_=self.U_conv3(torch.cat((d3,u3),1))
#print('u3_ shape:{}'.format(u3_.size()))
u2=self.U2(u3_)
u2=u2.contiguous()
#print('u2 shape:{}'.format(u2.size()))
u2_=self.U_conv2(torch.cat((d2,u2),1))
#print('u2_ shape:{}'.format(u2_.size()))
u1=self.U1(u2_)
u1=u1.contiguous()
#print('u1 shape:{}'.format(u1.size()))
u1_=self.U_conv1(torch.cat((d1,u1),1))
#print('u1_ shape:{}'.format(u1_.size()))
y=self.last(u1_)
#print('y shape:{}'.format(y.size()))
#y shape:(batch_size,channel,h,w)
return y
def Unet_test():
#测试程序
x=torch.randn(1,3,100,100)
model=Unet()
#print(model)
y=model(x)
original_size=list(x.size())
output_size=list(y.size())
print('the size of Unet model output is {}'.format(output_size))
assert (original_size[2]==output_size[2])&(original_size[3]==output_size[3]),\
"the size of output tensor:'{}' is wrong".format(original_size)
print('test pass!')
if __name__=='__main__':
Unet_test()