-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathhyper_net.py
311 lines (269 loc) · 11 KB
/
hyper_net.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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
# -*- coding: utf-8 -*-
"""
@author RuoyuLiu
@Email [email protected]
@这段说自己是author其实不太合适......这些网络架构有的我是根据人家论文里面的网络图仿真的，有的（比如ResNet-18这种烂大街网
络）就直接从网上找来改了改参数就放进来了，其实与其说自己是author不如说自己是一个mover。感谢各位作者的无私开源，之后我
会整理这些网络的思路来源，贴在下面。
"""
import torch
from torch import nn
from torch.nn import init
import torch.nn.functional as F
# 1.HResNet,快速好用
class HResNet(nn.Module):
def __init__(self, num_of_bands, num_of_class, patch_size):
super(HResNet, self).__init__()
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.conv0 = nn.Conv2d(self.num_of_bands, 64, kernel_size=(3, 3), padding=(0, 0))
self.bn1 = nn.BatchNorm2d(64)
self.relu11 = nn.ReLU()
self.conv11 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.relu12 = nn.ReLU()
self.conv12 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.bn2 = nn.BatchNorm2d(64)
self.relu21 = nn.ReLU()
self.conv21 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.relu22 = nn.ReLU()
self.conv22 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.avg_pool = nn.AvgPool2d((patch_size - 2, patch_size - 2))
# self.dense = nn.Linear(64 * (patch_size - 2) * (patch_size - 2), num_of_class)
self.dense = nn.Linear(64, num_of_class)
def forward(self, x):
x1 = self.conv0(x)
x0 = x1
x1 = self.bn1(x1)
x1 = self.relu11(x1)
x1 = self.conv11(x1)
x1 = self.relu12(x1)
x1 = self.conv12(x1)
x1 = x0 + x1
x2 = self.bn2(x1)
x2 = self.relu21(x2)
x2 = self.conv21(x2)
x2 = self.relu22(x2)
x2 = self.conv22(x2)
res = x1 + x2
# res = x0 + x1
res = self.avg_pool(res)
res = res.contiguous().view(res.size(0), -1)
res = self.dense(res)
return res
# 2.ResNet18及其残差模块
class ResidualBlock(nn.Module):
def __init__(self, inchannel, outchannel, stride=(1, 1)):
super(ResidualBlock, self).__init__()
self.left = nn.Sequential(
nn.Conv2d(inchannel, outchannel, kernel_size=(3, 3), stride=stride, padding=(1, 1), bias=False),
nn.BatchNorm2d(outchannel),
nn.ReLU(inplace=True),
nn.Conv2d(outchannel, outchannel, kernel_size=(3, 3), stride=stride, padding=(1, 1), bias=False),
nn.BatchNorm2d(outchannel)
)
self.shortcut = nn.Sequential()
if stride != 1 or inchannel != outchannel:
self.shortcut = nn.Sequential(
nn.Conv2d(inchannel, outchannel, kernel_size=(1, 1), stride=stride, bias=False),
nn.BatchNorm2d(outchannel)
)
def forward(self, x):
out = self.left(x)
out += self.shortcut(x)
out = F.relu(out)
return out
class ResNet(nn.Module):
def __init__(self, ResidualBlock, num_of_bands, num_of_class):
super(ResNet, self).__init__()
self.inchannel = 64
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.conv1 = nn.Sequential(
nn.Conv2d(self.num_of_bands, 64, kernel_size=(3, 3), padding=(0, 0), bias=False),
nn.BatchNorm2d(64),
nn.ReLU(),
)
self.layer1 = self.make_layer(ResidualBlock, 64, 2, stride=1)
self.layer2 = self.make_layer(ResidualBlock, 128, 2, stride=1)
self.layer3 = self.make_layer(ResidualBlock, 256, 2, stride=1)
self.layer4 = self.make_layer(ResidualBlock, 512, 2, stride=1)
self.fc = nn.Linear(512*7*7, self.num_of_class)
def make_layer(self, block, channels, num_blocks, stride):
strides = [stride] + [1] * (num_blocks - 1) # strides=[1,1]
layers = []
for stride in strides:
layers.append(block(self.inchannel, channels, stride))
self.inchannel = channels
return nn.Sequential(*layers)
def forward(self, x):
out = self.conv1(x)
out = self.layer1(out)
out = self.layer2(out)
out = self.layer3(out)
out = self.layer4(out)
# out = F.avg_pool2d(out, 4)
out = out.contiguous().view(out.size(0), -1)
out = self.fc(out)
return out
def ResNet18(num_of_bands, num_of_class):
return ResNet(ResidualBlock, num_of_bands, num_of_class)
# 3.典中典之2D-CNN
class CNN2d(nn.Module):
def __init__(self, num_of_bands, num_of_class, patch_size):
super(CNN2d, self).__init__()
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.conv00 = nn.Conv2d(self.num_of_bands, 64, kernel_size=(3, 3), padding=(0, 0))
self.conv01 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.conv02 = nn.Conv2d(64, 64, kernel_size=(3, 3), padding=(1, 1))
self.relu0 = nn.ReLU()
self.conv10 = nn.Conv2d(64, 128, kernel_size=(3, 3), padding=(0, 0))
self.conv11 = nn.Conv2d(128, 128, kernel_size=(3, 3), padding=(1, 1))
self.conv12 = nn.Conv2d(128, 128, kernel_size=(3, 3), padding=(1, 1))
self.relu1 = nn.ReLU()
self.conv20 = nn.Conv2d(128, 256, kernel_size=(3, 3), padding=(0, 0))
self.conv21 = nn.Conv2d(256, 256, kernel_size=(3, 3), padding=(1, 1))
self.conv22 = nn.Conv2d(256, 256, kernel_size=(1, 1), padding=(0, 0))
self.relu2 = nn.ReLU()
self.avg_pool = nn.AvgPool2d((patch_size - 6, patch_size - 6))
# self.dense = nn.Linear(256 * (patch_size - 6) * (patch_size - 6), num_of_class)
self.dense = nn.Linear(256, num_of_class)
def forward(self, x):
x0 = self.conv00(x)
x = self.conv00(x)
x0 = self.conv01(x0)
x0 = self.conv02(x0)
x0 = self.relu0(x0)
x = x0 + x
x1 = self.conv10(x)
x = self.conv10(x)
x1 = self.conv11(x1)
x1 = self.conv12(x1)
x1 = self.relu0(x1)
x = x1 + x
x2 = self.conv20(x)
x = self.conv20(x)
x2 = self.conv21(x2)
x2 = self.conv22(x2)
x2 = self.relu0(x2)
# out = x2 + x
res = x2 + x
out = self.avg_pool(res)
out = out.contiguous().view(out.size(0), -1)
out = self.dense(out)
return out
class FAST3DCNN(nn.Module):
def __init__(self, num_of_bands, num_of_class, patch_size):
super(FAST3DCNN, self).__init__()
self.patch_size = patch_size
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.conv1 = nn.Conv3d(1, 8, (7, 3, 3), padding=(0, 0, 0))
self.conv1_bn = nn.BatchNorm3d(8)
self.conv2 = nn.Conv3d(8, 16, (5, 3, 3), stride=(1, 1, 1), padding=(0, 0, 0))
self.conv2_bn = nn.BatchNorm3d(16)
self.conv3 = nn.Conv3d(16, 32, (3, 3, 3), stride=(1, 1, 1), padding=(0, 0, 0))
self.conv3_bn = nn.BatchNorm3d(32)
self.conv4 = nn.Conv3d(32, 64, (3, 3, 3), stride=(1, 1, 1), padding=(0, 0, 0))
self.conv4_bn = nn.BatchNorm3d(64)
self.dropout = nn.Dropout(p=0.4)
self.features_size = self._get_final_flattened_size()
self.fc1 = nn.Linear(self.features_size, 256)
self.fc2 = nn.Linear(256, 128)
self.fc3 = nn.Linear(128, self.num_of_class)
def _get_final_flattened_size(self):
with torch.no_grad():
x = torch.zeros((1, 1, self.num_of_bands, self.patch_size, self.patch_size))
x = self.conv1(x)
x = self.conv2(x)
x = self.conv3(x)
x = self.conv4(x)
b, t, c, w, h = x.size()
return b * t * c * w * h
def forward(self, x):
x = x.unsqueeze(1)
x = F.relu(self.conv1_bn(self.conv1(x)))
x = F.relu(self.conv2_bn(self.conv2(x)))
x = F.relu(self.conv3_bn(self.conv3(x)))
x = F.relu(self.conv4_bn(self.conv4(x)))
x = x.view(-1, self.features_size)
x = self.fc1(x)
x = self.dropout(x)
x = F.relu(x)
x = self.fc2(x)
x = self.dropout(x)
x = F.relu(x)
x = self.fc3(x)
return x
# # 5.DCGAN的生成器和判别器
class DCGenerator(nn.Module):
# initializers
def __init__(self, num_of_bands, num_of_class, noise_dim=17):
super(DCGenerator, self).__init__()
self.noise_dim = noise_dim
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.dense1 = nn.Linear(17, 100)
self.dense2 = nn.Linear(100, 3 * 3 * 512)
self.bn1 = nn.BatchNorm2d(momentum=0.8, num_features=512)
self.deconv1 = nn.ConvTranspose2d(512, 256, (3, 3), (1, 1))
self.relu1 = nn.ReLU()
self.bn2 = nn.BatchNorm2d(momentum=0.8, num_features=256)
self.deconv2 = nn.ConvTranspose2d(256, 128, (3, 3), (1, 1))
self.relu2 = nn.ReLU()
self.bn3 = nn.BatchNorm2d(momentum=0.8, num_features=128)
self.deconv3 = nn.ConvTranspose2d(128, num_of_bands, (3, 3), (1, 1))
self.tanh = nn.Tanh()
# forward method
def forward(self, inputs):
x = inputs.contiguous().view(inputs.size(0), -1)
x = self.dense1(x)
x = self.dense2(x)
x = x.contiguous().view(x.size(0), 512, 3, 3)
x = self.bn1(x)
x = self.deconv1(x)
x = self.relu1(x)
x = self.bn2(x)
x = self.deconv2(x)
x = self.relu2(x)
# x = self.bn3(x)
x = self.deconv3(x)
x = self.tanh(x)
return x
class DCDiscriminator(nn.Module):
# initializers
def __init__(self, num_of_bands, num_of_class):
super(DCDiscriminator, self).__init__()
self.num_of_bands = num_of_bands
self.num_of_class = num_of_class
self.conv1 = nn.Conv2d(num_of_bands, 128, (3, 3), (1, 1))
self.bn1 = nn.BatchNorm2d(128, momentum=0.8)
self.relu1 = nn.LeakyReLU()
self.conv2 = nn.Conv2d(128, 256, (3, 3), (1, 1))
self.bn2 = nn.BatchNorm2d(256, momentum=0.8)
self.relu2 = nn.LeakyReLU()
self.conv3 = nn.Conv2d(256, 512, (3, 3), (1, 1))
self.bn3 = nn.BatchNorm2d(512, momentum=0.8)
self.relu3 = nn.LeakyReLU()
self.conv4 = nn.Conv2d(512, 64, (3, 3), (1, 1))
self.bn4 = nn.BatchNorm2d(64, momentum=0.8)
self.dense_bin = nn.Linear(64, 1)
self.sigmoid = nn.Sigmoid()
self.dense_cls = nn.Linear(64, num_of_class)
def forward(self, inputs):
x = self.conv1(inputs)
x = self.bn1(x)
x = self.relu1(x)
x = self.conv2(x)
x = self.bn2(x)
x = self.relu2(x)
x = self.conv3(x)
x = self.bn3(x)
x = self.relu3(x)
x = self.conv4(x)
x = self.bn4(x)
x = x.contiguous().view(x.size(0), -1)
out_bin = self.dense_bin(x)
out_bin = self.sigmoid(out_bin)
out_cls = self.dense_cls(x)
return out_bin, out_cls