-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathDFFN.py
312 lines (227 loc) · 9.63 KB
/
DFFN.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
"""DFFN networks for HSI classification.
Based on:
Hyperspectral Image Classification With Deep Feature Fusion Network
Song, Li, Fang, Lu
https://github.com/weiweisong415/Demo_DFFN
Spectral size:
IP=3, PaviaU=5, Salinas=10
Spatial size:
IP=25, PaviaU=23, Salinas=27
Batch size: 100
"""
import tensorflow as tf
import pdb
def block1(inp, nchannels=16):
"""
out = in->Conv->BN->Relu->Conv->BN
out += in
"""
conv1 = tf.layers.conv2d(inp, nchannels, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
conv2 = tf.layers.conv2d(conv1, nchannels, 3, activation=None, padding='same')
conv2 = tf.layers.batch_normalization(conv2)
return tf.math.add(inp, conv2)
def block2(inp, nchannels=32):
"""Shrinks the input by 2
path1 = in -> strided conv1 -> bn
path2 = in -> strided conv3 -> bn->relu->conv3->bn
out = path1 + path2
"""
path1 = tf.layers.conv2d(inp, nchannels, 1, 2, activation=None, padding='same')
path1 = tf.layers.batch_normalization(path1)
path2 = tf.layers.conv2d(inp, nchannels, 3, 2, activation=None, padding='same')
path2 = tf.layers.batch_normalization(path2)
path2 = tf.nn.relu(path2)
path2 = tf.layers.conv2d(path2, nchannels, 3, activation=None, padding='same')
path2 = tf.layers.batch_normalization(path2)
return tf.math.add(path1, path2)
def DFFN_3tower_5depth(x_dict, dropout, reuse, is_training, n_classes):
"""Three towers. Each depth 5.
This is the train_paviaU network when input is 23.
20 steps/second.
94.5% on PaviaU 2%, lr 5e-5, within 10k
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
t2 = tf.nn.relu(block2(b1,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t3 = block2(t2,64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
# combine
t1out = tf.layers.conv2d(b1, 64, 3, 4, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
t2out = tf.layers.conv2d(t2, 64, 3, 2, activation=None, padding='same')
t2out = tf.layers.batch_normalization(t2out)
fuse = tf.math.add(t1out, t2out)
fuse = tf.math.add(fuse, t3)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))
def DFFN_3tower_4depth(x_dict, dropout, reuse, is_training, n_classes):
"""Three towers. Each depth 4.
This is the train_indian_pines network when input is 25.
This is the train_salinas network when input is 27.
23 steps/second.
97.0% IP p10, lr 5e-5
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
t2 = tf.nn.relu(block2(b1,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t3 = block2(t2,64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
# combine
t1out = tf.layers.conv2d(b1, 64, 3, 4, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
t2out = tf.layers.conv2d(t2, 64, 3, 2, activation=None, padding='same')
t2out = tf.layers.batch_normalization(t2out)
fuse = tf.math.add(t1out, t2out)
fuse = tf.math.add(fuse, t3)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))
def DFFN_3tower_3depth(x_dict, dropout, reuse, is_training, n_classes):
"""Three towers. Each depth 3.
30 steps/second.
Input as 25.
97.0% IP p10, within 10k itr, lr 5e-5
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
b1 = tf.nn.relu(block1(b1))
b1 = tf.nn.relu(block1(b1))
t2 = tf.nn.relu(block2(b1,32))
t2 = tf.nn.relu(block1(t2,32))
t2 = tf.nn.relu(block1(t2,32))
t3 = block2(t2,64)
t3 = block1(tf.nn.relu(t3),64)
t3 = block1(tf.nn.relu(t3),64)
# combine
t1out = tf.layers.conv2d(b1, 64, 3, 4, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
t2out = tf.layers.conv2d(t2, 64, 3, 2, activation=None, padding='same')
t2out = tf.layers.batch_normalization(t2out)
fuse = tf.math.add(t1out, t2out)
fuse = tf.math.add(fuse, t3)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))
def DFFN_3tower_2depth(x_dict, dropout, reuse, is_training, n_classes):
"""Three towers. Each depth 2.
40 steps/second.
Input as 25.
97.7% IP p10, within 10k itr, lr 1e-4 (should try lower lr)
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
b1 = tf.nn.relu(block1(b1))
t2 = tf.nn.relu(block2(b1,32))
t2 = tf.nn.relu(block1(t2,32))
t3 = block2(t2,64)
t3 = block1(tf.nn.relu(t3),64)
# combine
t1out = tf.layers.conv2d(b1, 64, 3, 4, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
t2out = tf.layers.conv2d(t2, 64, 3, 2, activation=None, padding='same')
t2out = tf.layers.batch_normalization(t2out)
fuse = tf.math.add(t1out, t2out)
fuse = tf.math.add(fuse, t3)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))
def DFFN_3tower_1depth(x_dict, dropout, reuse, is_training, n_classes):
"""Three towers. Each depth 1.
60 steps/second.
Input as 25.
~90% IP p05, within 4k itr, lr 1e-4
96.8% IP p10, within 10k itr, lr 1e-4
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
t2 = tf.nn.relu(block2(b1,32))
t3 = block2(t2,64)
# combine
t1out = tf.layers.conv2d(b1, 64, 3, 4, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
t2out = tf.layers.conv2d(t2, 64, 3, 2, activation=None, padding='same')
t2out = tf.layers.batch_normalization(t2out)
fuse = tf.math.add(t1out, t2out)
fuse = tf.math.add(fuse, t3)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))
def DFFN_2tower_1depth(x_dict, dropout, reuse, is_training, n_classes):
"""Two towers. Each depth 1.
Input as 13.
84% IP p05, within 4k itr, lr 1e-2
"""
with tf.variable_scope('DFFN', reuse=reuse):
x = x_dict['subimages']
conv1 = tf.layers.conv2d(x, 16, 3, activation=None, padding='same')
conv1 = tf.layers.batch_normalization(conv1)
conv1 = tf.nn.relu(conv1)
# beginning
b1 = tf.nn.relu(block1(conv1))
t2 = block2(b1,32)
# combine
t1out = tf.layers.conv2d(b1, 32, 3, 2, activation=None, padding='same')
t1out = tf.layers.batch_normalization(t1out)
fuse = tf.math.add(t2, t1out)
fuse = tf.nn.relu(fuse)
# end
fuse = tf.reduce_mean(fuse, axis=(1,2), keepdims=True) # avg pooling
fuse = tf.layers.dense(fuse, n_classes)
return tf.squeeze(fuse, axis=(1,2))