semantic_segmentation_implementation/deconvnet.py at main · tjdtnsu/semantic_segmentation_implementation · 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
148
import torch
from torch import nn
from torch.nn import functional as F

class ConvLayer1(nn.Module):
    def __init__(self, channel_input, channel_output):
        super(ConvLayer1, self).__init__()
        self.conv1 = nn.Conv2d(channel_input, channel_output, 3, padding=1)
        self.batch1 = nn.BatchNorm2d(channel_output)
        self.conv2 = nn.Conv2d(channel_output, channel_output, 3, padding=1)
        self.batch2 = nn.BatchNorm2d(channel_output)
        self.pool = nn.MaxPool2d(2, return_indices=True)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.batch1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.batch2(x)
        x = self.relu(x)
        return self.pool(x)


class ConvLayer2(nn.Module):
    def __init__(self, channel_input, channel_output):
        super(ConvLayer2, self).__init__()
        self.conv1 = nn.Conv2d(channel_input, channel_output, 3, padding=1)
        self.batch1 = nn.BatchNorm2d(channel_output)
        self.conv2 = nn.Conv2d(channel_output, channel_output, 3, padding=1)
        self.batch2 = nn.BatchNorm2d(channel_output)
        self.conv3 = nn.Conv2d(channel_output, channel_output, 3, padding=1)
        self.batch3 = nn.BatchNorm2d(channel_output)
        self.pool = nn.MaxPool2d(2, return_indices=True)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.batch1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.batch2(x)
        x = self.relu(x)
        x = self.conv3(x)
        x = self.batch3(x)
        x = self.relu(x)
        return self.pool(x)


class DeconvLayer1(nn.Module):
    def __init__(self, channel_input, channel_output):
        super(DeconvLayer1, self).__init__()
        self.unpool = nn.MaxUnpool2d(2)
        self.deconv1 = nn.ConvTranspose2d(channel_input, channel_input, 3, padding=1)
        self.batch1 = nn.BatchNorm2d(channel_input)
        self.deconv2 = nn.ConvTranspose2d(channel_input, channel_output, 3, padding=1)
        self.batch2 = nn.BatchNorm2d(channel_output)
        self.relu = nn.ReLU()

    def forward(self, x, indicate):
        x = self.unpool(x, indicate)
        x = self.deconv1(x)
        x = self.batch1(x)
        x = self.relu(x)
        x = self.deconv2(x)
        x = self.batch2(x)
        return self.relu(x)


class DeconvLayer2(nn.Module):
    def __init__(self, channel_input, channel_output):
        super(DeconvLayer2, self).__init__()
        self.unpool = nn.MaxUnpool2d(2)
        self.deconv1 = nn.ConvTranspose2d(channel_input, channel_input, 3, padding=1)
        self.batch1 = nn.BatchNorm2d(channel_input)
        self.deconv2 = nn.ConvTranspose2d(channel_input, channel_input, 3, padding=1)
        self.batch2 = nn.BatchNorm2d(channel_input)
        self.deconv3 = nn.ConvTranspose2d(channel_input, channel_output, 3, padding=1)
        self.batch3 = nn.BatchNorm2d(channel_output)
        self.relu = nn.ReLU()

    def forward(self, x, indicate):
        x = self.unpool(x, indicate)
        x = self.deconv1(x)
        x = self.batch1(x)
        x = self.relu(x)
        x = self.deconv2(x)
        x = self.batch2(x)
        x = self.relu(x)
        x = self.deconv3(x)
        x = self.batch3(x)
        return self.relu(x)


class FCLayer(nn.Module):
    def __init__(self):
        super(FCLayer, self).__init__()
        self.fc1 = nn.Conv2d(512, 4096, 7, padding=0)
        self.batch1 = nn.BatchNorm2d(4096)
        self.fc2 = nn.Conv2d(4096, 4096, 1, padding=0)
        self.batch2 = nn.BatchNorm2d(4096)
        self.fc3 = nn.ConvTranspose2d(4096, 512, 7)
        self.batch3 = nn.BatchNorm2d(512)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.fc1(x)
        x = self.batch1(x)
        x = self.relu(x)
        x = self.fc2(x)
        x = self.batch2(x)
        x = self.relu(x)
        x = self.fc3(x)
        x = self.batch3(x)
        return self.relu(x)


# input_size = (224, 224, 3)

class DeconvNet(nn.Module):
    def __init__(self, n_class):
        super(DeconvNet, self).__init__()
        self.convLayer1 = ConvLayer1(3, 64)
        self.convLayer2 = ConvLayer1(64, 128)
        self.convLayer3 = ConvLayer2(128, 256)
        self.convLayer4 = ConvLayer2(256, 512)
        self.convLayer5 = ConvLayer2(512, 512)
        self.fcLayer = FCLayer()
        self.deconvLayer5 = DeconvLayer2(512, 512)
        self.deconvLayer4 = DeconvLayer2(512, 256)
        self.deconvLayer3 = DeconvLayer2(256, 128)
        self.deconvLayer2 = DeconvLayer1(128, 64)
        self.deconvLayer1 = DeconvLayer1(64, 64)
        self.segScore = nn.Conv2d(64, n_class, 1, padding=0)

    def forward(self, x):
        x, i1 = self.convLayer1(x)
        x, i2 = self.convLayer2(x)
        x, i3 = self.convLayer3(x)
        x, i4 = self.convLayer4(x)
        x, i5 = self.convLayer5(x)
        x = self.fcLayer(x)
        x = self.deconvLayer5(x, i5)
        x = self.deconvLayer4(x, i4)
        x = self.deconvLayer3(x, i3)
        x = self.deconvLayer2(x, i2)
        x = self.deconvLayer1(x, i1)
        return self.segScore(x)