-
Notifications
You must be signed in to change notification settings - Fork 3
/
AUIFNet.py
120 lines (111 loc) · 4.23 KB
/
AUIFNet.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
# -*- coding: utf-8 -*-
"""
@author: Zixiang Zhao (zixiangzhao@stu.xjtu.edu.cn)
Pytorch implement for "Efficient and Model-Based Infrared and Visible Image Fusion via Algorithm Unrolling" (TCSVT 2021)
10.1109/TCSVT.2021.3075745
https://ieeexplore.ieee.org/document/9416456
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from args import channel,img_size,layer_numb
import kornia
#定义初始化滤波
Laplace = kornia.filters.Laplacian(19)#高通滤波
Blur = kornia.filters.BoxBlur((11, 11))#低通滤波
# =============================================================================
# 网络搭建
# =============================================================================
class BCL(nn.Module):
def __init__(self):
super(BCL, self).__init__()
self.kernel = nn.Parameter(torch.randn(size=[64, 1, 3, 3]))
self.zero_pad=nn.ReflectionPad2d(1)
self.rest=nn.Sequential(
nn.BatchNorm2d(1),
nn.PReLU()
)
self.eta=nn.Parameter(
nn.init.normal_(
torch.empty(1).cuda(),mean=0.1,std=0.03
))
self.theta=nn.Parameter(
nn.init.normal_(
torch.empty(1).cuda(),mean=1e-3,std=1e-4
))
def forward(self,x_in,img):
x_in_2=self.zero_pad(x_in)
x_1=F.conv2d(x_in_2,self.kernel,padding=0)
x_2=self.zero_pad(x_1)
kernel2 = torch.rot90(self.kernel, 2, [-1,-2]).transpose(0,1)
x_3=F.conv2d(x_2,kernel2,padding=0)
x_out=self.rest(x_3)
x_out_2=x_in-self.eta*(x_out-self.theta*(img-x_in))
return x_out_2,img,self.eta.item(),self.theta.item()
class DCL(nn.Module):
def __init__(self):
super(DCL, self).__init__()
self.kernel = nn.Parameter(torch.randn(size=[64, 1, 3, 3]))
self.zero_pad=nn.ReflectionPad2d(1)
self.rest=nn.Sequential(
nn.BatchNorm2d(1),
nn.PReLU()
)
self.eta=nn.Parameter(
nn.init.normal_(
torch.empty(1).cuda(),mean=0.1,std=0.03
))
self.theta=nn.Parameter(
nn.init.normal_(
torch.empty(1).cuda(),mean=1e-3,std=1e-4
))
def forward(self,x_in,img):
x_in_2=self.zero_pad(x_in)
x_1=F.conv2d(x_in_2,self.kernel,padding=0)
x_2=self.zero_pad(x_1)
kernel2 = torch.rot90(self.kernel, 2, [-1,-2]).transpose(0,1)
x_3=F.conv2d(x_2,kernel2,padding=0)
x_out=self.rest(x_3)
x_out_2=x_in-self.eta*(x_out-self.theta*(img-x_in))
return x_out_2,img,self.eta.item(),self.theta.item()
class Encoder_Base(nn.Module):
def __init__(self,size=img_size):
super(Encoder_Base, self).__init__()
self.numb=layer_numb
self.conv1 = nn.ModuleList([BCL() for i in range(self.numb)])
def forward(self,img):
img_blur=Blur(img)
eta_list_base=[]
theta_list_base=[]
for layer in self.conv1:
img_blur, img, eta, theta = layer(img_blur,img)
eta_list_base.append(eta)
theta_list_base.append(theta)
return img_blur,eta_list_base,theta_list_base
class Encoder_Detail(nn.Module):
def __init__(self,size=img_size):
super(Encoder_Detail, self).__init__()
self.numb=layer_numb
self.conv2 = nn.ModuleList([DCL() for i in range(self.numb)])
def forward(self,img):
img_laplace=Laplace(img)
eta_list_detail=[]
theta_list_detail=[]
for layer in self.conv2:
img_laplace, img, eta, theta = layer(img_laplace,img)
eta_list_detail.append(eta)
theta_list_detail.append(theta)
return img_laplace,eta_list_detail,theta_list_detail
class Decoder(nn.Module):
def __init__(self):
super(Decoder, self).__init__()
self.decoder = nn.Sequential(
nn.ReflectionPad2d(1),
nn.Conv2d(1, 1, 3, padding=0, bias=False), # in_channels, out_channels, kernel_size
nn.BatchNorm2d(1),
nn.Sigmoid()
)
def forward(self, fm_b, fm_d):
fm=fm_b+fm_d
return self.decoder(fm)