LCNet/LCNet_model.py at main · shuowenyang/LCNet · GitHub

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
import torch.nn as nn
import torch
import torch.nn.functional as F
from torch.nn import init


class ResBlock(nn.Module):
    def __init__(self, dim):
        super(ResBlock, self).__init__()

        self.conv1 = nn.Conv2d(dim, dim, kernel_size=3, padding=1, bias=True)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(dim, dim, kernel_size=3, padding=1, bias=True)

    def forward(self, x):
        return x + self.conv2(self.relu(self.conv1(x)))


class FeaMod1(nn.Module):
    def __init__(self, dim):
        super(FeaMod1, self).__init__()

        self.res1 = ResBlock(dim)

    def forward(self, x):
        x = self.res1(x)

        return x


class FeaMod2(nn.Module):
    def __init__(self, dim):
        super(FeaMod2, self).__init__()

        self.dowm = nn.Conv2d(dim, dim * 2, kernel_size=3, padding=1, stride=2, bias=True)
        self.res1 = ResBlock(dim * 2)

    def forward(self, x):
        x = self.res1(self.dowm(x))
        x = F.interpolate(x, scale_factor=2, mode='bilinear')

        return x


class FeaMod(nn.Module):
    def __init__(self, dim):
        super(FeaMod, self).__init__()

        self.fea1 = FeaMod1(dim)
        self.fea2 = FeaMod2(dim)
        self.conv1 = nn.Conv2d(dim * 3, dim, kernel_size=1, padding=0, stride=1, bias=True)

    def forward(self, x):
        x1 = self.fea1(x)
        x2 = self.fea2(x)
        x = torch.cat((x1, x2), dim=1)
        x = self.conv1(x)

        return x


class IgMod(nn.Module):
    def __init__(self, dim):
        super(IgMod, self).__init__()

        self.conv1 = nn.Conv2d(dim, 1, kernel_size=3, padding=1, bias=True)
        self.conv2 = nn.Conv2d(1, dim, kernel_size=3, padding=1, bias=True)
        self.res1 = ResBlock(dim)

    def forward(self, x, y, Phi, PhiT):
        x_pixel = self.conv1(x)
        Phix = F.conv2d(x_pixel, Phi, padding=0, stride=32, bias=None)
        delta = y - Phix
        x_pixel = nn.PixelShuffle(32)(F.conv2d(delta, PhiT, padding=0, bias=None))
        x_delta = self.conv2(x_pixel)
        x = self.res1(x_delta) + x

        return x


class AttBlock(nn.Module):
    def __init__(self, dim):
        super(AttBlock, self).__init__()

        self.conv1 = nn.Conv2d(dim, dim, kernel_size=7, padding=3, stride=1, bias=True, groups=dim)
        self.conv2 = nn.Conv2d(dim, dim, kernel_size=1, padding=0, stride=1, bias=True)
        self.conv3 = nn.Conv2d(dim, dim, kernel_size=1, padding=0, stride=1, bias=True)
        self.scale = dim ** -0.5
        self.conv4 = nn.Conv2d(dim, dim, kernel_size=1, padding=0, stride=1, bias=True)

    def forward(self, x):
        B, C, H, W = x.shape
        x = self.conv1(x)
        x = x.reshape(B, C, H // 32, 32, W // 32, 32).permute(0, 2, 4, 1, 3, 5).reshape(-1, C, 32, 32)
        x1 = self.conv2(x).reshape(-1, C, 32 * 32)
        x2 = self.conv3(x).reshape(-1, C, 32 * 32).transpose(1, 2)
        att = (x2 @ x1) * self.scale
        att = att.softmax(dim=1)
        x = (x.reshape(-1, C, 32 * 32) @ att).reshape(-1, C, 32, 32)
        x = self.conv4(x)
        x = x.reshape(B, H // 32, W // 32, C, 32, 32).permute(0, 3, 1, 4, 2, 5).reshape(B, C, H, W)

        return x


class LCNet(nn.Module):
    def __init__(self, sensing_rate, LayerNo):
        super(LCNet, self).__init__()

        self.measurement = int(sensing_rate * 1024)
        self.base = 16

        self.Phi = nn.Parameter(init.xavier_normal_(torch.Tensor(self.measurement, 1024)))
        self.conv1 = nn.Conv2d(1, self.base, kernel_size=3, padding=1, bias=True)
        self.conv2 = nn.Conv2d(self.base, 2, kernel_size=1, padding=0, bias=True)
        layer1 = []
        layer2 = []
        self.LayerNo = LayerNo
        for i in range(LayerNo):
            layer1.append(FeaMod(self.base))
            layer2.append(IgMod(self.base))
        self.fcs1 = nn.ModuleList(layer1)
        self.fcs2 = nn.ModuleList(layer2)

    def forward(self, x):

        Phi = self.Phi.contiguous().view(self.measurement, 1, 32, 32)#######102,1,32,32

        PhiT = self.Phi.t().contiguous().view(1024, self.measurement, 1, 1)

        y = F.conv2d(x, Phi, padding=0, stride=32, bias=None)##########1,102,8,8

        x = F.conv2d(y, PhiT, padding=0, bias=None)############1,1024,8,8

        x = nn.PixelShuffle(32)(x)######1,1,256,256

        x = self.conv1(x)##########1,16,256,256

        for i in range(self.LayerNo):
            x = self.fcs2[i](x, y, Phi, PhiT)
            x = self.fcs1[i](x)#############1,16,256,256

        x = self.conv2(x)############,1,2,256,256

        phi_cons = torch.mm(self.Phi, self.Phi.t())

        return x, phi_cons,y,Phi