-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathMACUNet.py
230 lines (188 loc) · 8.48 KB
/
MACUNet.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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
import torch
from torch import nn
class ChannelAttention(nn.Module):
def __init__(self, in_planes, out_planes, ratio=2):
super(ChannelAttention, self).__init__()
self.conv = nn.Conv2d(in_planes, out_planes, 1, bias=False)
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.max_pool = nn.AdaptiveMaxPool2d(1)
self.fc11 = nn.Conv2d(out_planes, out_planes // ratio, 1, bias=False)
self.fc12 = nn.Conv2d(out_planes // ratio, out_planes, 1, bias=False)
self.fc21 = nn.Conv2d(out_planes, out_planes // ratio, 1, bias=False)
self.fc22 = nn.Conv2d(out_planes // ratio, out_planes, 1, bias=False)
self.relu1 = nn.ReLU(True)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
x = self.conv(x)
avg_out = self.fc12(self.relu1(self.fc11(self.avg_pool(x))))
max_out = self.fc22(self.relu1(self.fc21(self.max_pool(x))))
out = avg_out + max_out
del avg_out, max_out
return x * self.sigmoid(out)
def conv3otherRelu(in_planes, out_planes, kernel_size=None, stride=None, padding=None):
# 3x3 convolution with padding and relu
if kernel_size is None:
kernel_size = 3
assert isinstance(kernel_size, (int, tuple)), 'kernel_size is not in (int, tuple)!'
if stride is None:
stride = 1
assert isinstance(stride, (int, tuple)), 'stride is not in (int, tuple)!'
if padding is None:
padding = 1
assert isinstance(padding, (int, tuple)), 'padding is not in (int, tuple)!'
return nn.Sequential(
nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding, bias=True),
# nn.ReLU() # inplace=True
nn.LeakyReLU()
)
class ACBlock(nn.Module):
def __init__(self, in_planes, out_planes):
super(ACBlock, self).__init__()
self.squre = nn.Conv2d(in_planes, out_planes, kernel_size=3, padding=1, stride=1)
self.cross_ver = nn.Conv2d(in_planes, out_planes, kernel_size=(1, 3), padding=(0, 1), stride=1)
self.cross_hor = nn.Conv2d(in_planes, out_planes, kernel_size=(3, 1), padding=(1, 0), stride=1)
self.bn = nn.BatchNorm2d(out_planes)
self.ReLU = nn.ReLU(True)
def forward(self, x):
x1 = self.squre(x)
x2 = self.cross_ver(x)
x3 = self.cross_hor(x)
return self.ReLU(self.bn(x1 + x2 + x3))
class MACUNet(nn.Module):
def __init__(self, band_num, class_num):
super(MACUNet, self).__init__()
self.band_num = band_num
self.class_num = class_num
self.name = 'MACUNet'
# channels = [32, 64, 128, 256, 512]
channels = [16, 32, 64, 128, 256, 512]
self.conv1 = nn.Sequential(
ACBlock(self.band_num, channels[0]),
ACBlock(channels[0], channels[0])
)
self.conv12 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[0], channels[1])
)
self.conv13 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[1], channels[2]),
)
self.conv14 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[2], channels[3])
)
self.conv2 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[0], channels[1]),
ACBlock(channels[1], channels[1])
)
self.conv23 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[1], channels[2])
)
self.conv24 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[2], channels[3])
)
self.conv3 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[1], channels[2]),
ACBlock(channels[2], channels[2]),
ACBlock(channels[2], channels[2])
)
self.conv34 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[2], channels[3])
)
self.conv4 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[2], channels[3]),
ACBlock(channels[3], channels[3]),
ACBlock(channels[3], channels[3])
)
self.conv5 = nn.Sequential(
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)),
ACBlock(channels[3], channels[4]),
ACBlock(channels[4], channels[4]),
ACBlock(channels[4], channels[4])
)
self.skblock4 = ChannelAttention(channels[3]*5, channels[3]*2, 16)
self.skblock3 = ChannelAttention(channels[2]*5, channels[2]*2, 16)
self.skblock2 = ChannelAttention(channels[1]*5, channels[1]*2, 16)
self.skblock1 = ChannelAttention(channels[0]*5, channels[0]*2, 16)
self.deconv4 = nn.ConvTranspose2d(channels[4], channels[3], kernel_size=(2, 2), stride=(2, 2))
self.deconv43 = nn.ConvTranspose2d(channels[3], channels[2], kernel_size=(2, 2), stride=(2, 2))
self.deconv42 = nn.ConvTranspose2d(channels[2], channels[1], kernel_size=(2, 2), stride=(2, 2))
self.deconv41 = nn.ConvTranspose2d(channels[1], channels[0], kernel_size=(2, 2), stride=(2, 2))
self.conv6 = nn.Sequential(
ACBlock(channels[4], channels[3]),
ACBlock(channels[3], channels[3]),
)
self.deconv3 = nn.ConvTranspose2d(channels[3], channels[2], kernel_size=(2, 2), stride=(2, 2))
self.deconv32 = nn.ConvTranspose2d(channels[2], channels[1], kernel_size=(2, 2), stride=(2, 2))
self.deconv31 = nn.ConvTranspose2d(channels[1], channels[0], kernel_size=(2, 2), stride=(2, 2))
self.conv7 = nn.Sequential(
ACBlock(channels[3], channels[2]),
ACBlock(channels[2], channels[2]),
)
self.deconv2 = nn.ConvTranspose2d(channels[2], channels[1], kernel_size=(2, 2), stride=(2, 2))
self.deconv21 = nn.ConvTranspose2d(channels[1], channels[0], kernel_size=(2, 2), stride=(2, 2))
self.conv8 = nn.Sequential(
ACBlock(channels[2], channels[1]),
ACBlock(channels[1], channels[1])
)
self.deconv1 = nn.ConvTranspose2d(channels[1], channels[0], kernel_size=(2, 2), stride=(2, 2))
self.conv9 = nn.Sequential(
ACBlock(channels[1], channels[0]),
ACBlock(channels[0], channels[0])
)
self.conv10 = nn.Conv2d(channels[0], self.class_num, kernel_size=1, stride=1)
def forward(self, x):
conv1 = self.conv1(x)
conv12 = self.conv12(conv1)
conv13 = self.conv13(conv12)
conv14 = self.conv14(conv13)
conv2 = self.conv2(conv1)
conv23 = self.conv23(conv2)
conv24 = self.conv24(conv23)
conv3 = self.conv3(conv2)
conv34 = self.conv34(conv3)
conv4 = self.conv4(conv3)
conv5 = self.conv5(conv4)
deconv4 = self.deconv4(conv5)
deconv43 = self.deconv43(deconv4)
deconv42 = self.deconv42(deconv43)
deconv41 = self.deconv41(deconv42)
conv6 = torch.cat((deconv4, conv4, conv34, conv24, conv14), 1)
conv6 = self.skblock4(conv6)
conv6 = self.conv6(conv6)
del deconv4, conv4, conv34, conv24, conv14, conv5
deconv3 = self.deconv3(conv6)
deconv32 = self.deconv32(deconv3)
deconv31 = self.deconv31(deconv32)
conv7 = torch.cat((deconv3, deconv43, conv3, conv23, conv13), 1)
conv7 = self.skblock3(conv7)
conv7 = self.conv7(conv7)
del deconv3, deconv43, conv3, conv23, conv13, conv6
deconv2 = self.deconv2(conv7)
deconv21 = self.deconv21(deconv2)
conv8 = torch.cat((deconv2, deconv42, deconv32, conv2, conv12), 1)
conv8 = self.skblock2(conv8)
conv8 = self.conv8(conv8)
del deconv2, deconv42, deconv32, conv2, conv12, conv7
deconv1 = self.deconv1(conv8)
conv9 = torch.cat((deconv1, deconv41, deconv31, deconv21, conv1), 1)
conv9 = self.skblock1(conv9)
conv9 = self.conv9(conv9)
# conv9 = self.seblock(conv9)
del deconv1, deconv41, deconv31, deconv21, conv1, conv8
output = self.conv10(conv9)
return output
if __name__ == '__main__':
num_classes = 10
in_batch, inchannel, in_h, in_w = 4, 3, 128, 128
x = torch.randn(in_batch, inchannel, in_h, in_w)
net = MACUNet(inchannel, num_classes)
out = net(x)
print(out.shape)