-
Notifications
You must be signed in to change notification settings - Fork 28
/
Copy path2_pytorch2keras.py
391 lines (274 loc) · 14 KB
/
2_pytorch2keras.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
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
#! /usr/bin/env python
# coding=utf-8
#================================================================
#
# Author : miemie2013
# Created date: 2020-01-11 16:31:57
# Description : pytorch_yolov3
# 导出为仓库 https://github.com/miemie2013/Keras-DIOU-YOLOv3
# 所使用的的Keras模型。
#
#================================================================
import torch
import keras
import keras.layers as layers
from model.darknet_yolo_pt import Darknet
import platform
print(torch.cuda.is_available())
sysstr = platform.system()
print(torch.__version__)
if sysstr == 'Windows':
torch.backends.cudnn.enabled = False
def conv2d_unit(x, filters, kernels, strides=1, padding='same'):
x = layers.Conv2D(filters, kernels,
padding=padding,
strides=strides,
use_bias=False,
activation='linear',
kernel_regularizer=keras.regularizers.l2(5e-4))(x)
x = layers.BatchNormalization()(x)
x = keras.layers.advanced_activations.LeakyReLU(alpha=0.1)(x)
return x
def residual_block(inputs, filters):
x = conv2d_unit(inputs, filters, (1, 1))
x = conv2d_unit(x, 2 * filters, (3, 3))
x = layers.add([inputs, x])
x = layers.Activation('linear')(x)
return x
def stack_residual_block(inputs, filters, n):
x = residual_block(inputs, filters)
for i in range(n - 1):
x = residual_block(x, filters)
return x
def find(base_model, conv2d_name, batch_normalization_name):
i1, i2 = -1, -1
for i in range(len(base_model.layers)):
if base_model.layers[i].name == conv2d_name:
i1 = i
if base_model.layers[i].name == batch_normalization_name:
i2 = i
return i1, i2
def aaaaaaa(conv, bn, cccccccc):
conv2, bn2 = cccccccc.conv, cccccccc.bn
w = conv2.weight.data.numpy()
y = bn2.weight.data.numpy()
b = bn2.bias.data.numpy()
m = bn2.running_mean.data.numpy()
v = bn2.running_var.data.numpy()
w = w.transpose(2, 3, 1, 0)
conv.set_weights([w])
bn.set_weights([y, b, m, v])
def aaaaaaa2(conv, cccccccc):
w = cccccccc.weight.data.numpy()
b = cccccccc.bias.data.numpy()
w = w.transpose(2, 3, 1, 0)
conv.set_weights([w, b])
def bbbbbbbbbbb(base_model, stack_residual_block, start_index):
i = start_index
for residual_block in stack_residual_block.sequential:
conv1 = residual_block.conv1
conv2 = residual_block.conv2
i1, i2 = find(base_model, 'conv2d_%d'%(i, ), 'batch_normalization_%d'%(i, ))
aaaaaaa(base_model.layers[i1], base_model.layers[i2], conv1)
i1, i2 = find(base_model, 'conv2d_%d'%(i+1, ), 'batch_normalization_%d'%(i+1, ))
aaaaaaa(base_model.layers[i1], base_model.layers[i2], conv2)
i += 2
if __name__ == '__main__':
num_classes = 20
self = Darknet(num_classes, initial_filters=8)
self.load_state_dict(torch.load('aaaa_bgr.pt'))
self.eval() # 必须调用model.eval()来设置dropout和batch normalization layers在运行推理前，切换到评估模式. 不这样做的化会产生不一致的推理结果.
# pattern取0时，初始卷积核个数
initial_filters = 8
i32 = initial_filters
i64 = i32 * 2
i128 = i32 * 4
i256 = i32 * 8
i512 = i32 * 16
i1024 = i32 * 32
# 多尺度训练
# inputs = layers.Input(shape=(None, None, 3))
inputs = layers.Input(shape=(416, 416, 3))
''' darknet53部分，所有卷积层都没有偏移use_bias=False '''
x = conv2d_unit(inputs, i32, (3, 3))
x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(x)
x = conv2d_unit(x, i64, (3, 3), strides=2, padding='valid')
x = stack_residual_block(x, i32, n=1)
x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(x)
x = conv2d_unit(x, i128, (3, 3), strides=2, padding='valid')
x = stack_residual_block(x, i64, n=2)
x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(x)
x = conv2d_unit(x, i256, (3, 3), strides=2, padding='valid')
act11 = stack_residual_block(x, i128, n=8)
x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(act11)
x = conv2d_unit(x, i512, (3, 3), strides=2, padding='valid')
act19 = stack_residual_block(x, i256, n=8)
x = layers.ZeroPadding2D(padding=((1, 0), (1, 0)))(act19)
x = conv2d_unit(x, i1024, (3, 3), strides=2, padding='valid')
act23 = stack_residual_block(x, i512, n=4)
''' darknet53部分结束，余下部分不再有残差块stack_residual_block() '''
''' 除了y1 y2 y3之前的1x1卷积有偏移，所有卷积层都没有偏移use_bias=False '''
x = conv2d_unit(act23, i512, (1, 1), strides=1)
x = conv2d_unit(x, i1024, (3, 3), strides=1)
x = conv2d_unit(x, i512, (1, 1), strides=1)
x = conv2d_unit(x, i1024, (3, 3), strides=1)
lkrelu57 = conv2d_unit(x, i512, (1, 1), strides=1)
x = conv2d_unit(lkrelu57, i1024, (3, 3), strides=1)
y1 = layers.Conv2D(3 * (num_classes + 5), (1, 1))(x)
# y1 = layers.Reshape((13, 13, 3, (num_classes + 5)), name='y1_r')(x)
x = conv2d_unit(lkrelu57, i256, (1, 1), strides=1)
x = layers.UpSampling2D(2)(x)
x = layers.Concatenate()([x, act19])
x = conv2d_unit(x, i256, (1, 1), strides=1)
x = conv2d_unit(x, i512, (3, 3), strides=1)
x = conv2d_unit(x, i256, (1, 1), strides=1)
x = conv2d_unit(x, i512, (3, 3), strides=1)
lkrelu64 = conv2d_unit(x, i256, (1, 1), strides=1)
x = conv2d_unit(lkrelu64, i512, (3, 3), strides=1)
y2 = layers.Conv2D(3 * (num_classes + 5), (1, 1))(x)
# y2 = layers.Reshape((26, 26, 3, (num_classes + 5)), name='y2_r')(x)
x = conv2d_unit(lkrelu64, i128, (1, 1), strides=1)
x = layers.UpSampling2D(2)(x)
x = layers.Concatenate()([x, act11])
x = conv2d_unit(x, i128, (1, 1), strides=1)
x = conv2d_unit(x, i256, (3, 3), strides=1)
x = conv2d_unit(x, i128, (1, 1), strides=1)
x = conv2d_unit(x, i256, (3, 3), strides=1)
x = conv2d_unit(x, i128, (1, 1), strides=1)
x = conv2d_unit(x, i256, (3, 3), strides=1)
y3 = layers.Conv2D(3 * (num_classes + 5), (1, 1))(x)
# y3 = layers.Reshape((52, 52, 3, (num_classes + 5)), name='y3_r')(x)
base_model = keras.models.Model(inputs=inputs, outputs=[y1, y2, y3])
base_model.summary()
# 复制参数
print()
i1, i2 = find(base_model, 'conv2d_1', 'batch_normalization_1')
# self.conv1 = Conv2dUnit(3, i32, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv1)
dd = 2
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv2 = Conv2dUnit(i32, i64, (3, 3), stride=2, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv2)
# self.stack_residual_block_1 = StackResidualBlock(i64, i32, n=1)
bbbbbbbbbbb(base_model, self.stack_residual_block_1, start_index=3)
dd = 5
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv3 = Conv2dUnit(i64, i128, (3, 3), stride=2, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv3)
# self.stack_residual_block_2 = StackResidualBlock(i128, i64, n=2)
bbbbbbbbbbb(base_model, self.stack_residual_block_2, start_index=6)
dd = 10
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv4 = Conv2dUnit(i128, i256, (3, 3), stride=2, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv4)
# self.stack_residual_block_3 = StackResidualBlock(i256, i128, n=8)
bbbbbbbbbbb(base_model, self.stack_residual_block_3, start_index=11)
dd = 27
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv5 = Conv2dUnit(i256, i512, (3, 3), stride=2, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv5)
# self.stack_residual_block_4 = StackResidualBlock(i512, i256, n=8)
bbbbbbbbbbb(base_model, self.stack_residual_block_4, start_index=28)
dd = 44
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv6 = Conv2dUnit(i512, i1024, (3, 3), stride=2, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv6)
# self.stack_residual_block_5 = StackResidualBlock(i1024, i512, n=4)
bbbbbbbbbbb(base_model, self.stack_residual_block_5, start_index=45)
''' darknet53部分结束 '''
dd = 53
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv53 = Conv2dUnit(i1024, i512, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv53)
dd = 54
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv54 = Conv2dUnit(i512, i1024, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv54)
dd = 55
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv55 = Conv2dUnit(i1024, i512, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv55)
dd = 56
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv56 = Conv2dUnit(i512, i1024, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv56)
dd = 57
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv57 = Conv2dUnit(i1024, i512, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv57)
dd = 58
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv58 = Conv2dUnit(i512, i1024, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv58)
dd = 59
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv59 = torch.nn.Conv2d(i1024, 3*(num_classes + 5), kernel_size=(1, 1))
aaaaaaa2(base_model.layers[i1], self.conv59)
dd = 60
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv60 = Conv2dUnit(i512, i256, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv60)
# self.upsample1 = torch.nn.Upsample(scale_factor=2, mode='nearest')
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv61 = Conv2dUnit(i256+i512, i256, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv61)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv62 = Conv2dUnit(i256, i512, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv62)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv63 = Conv2dUnit(i512, i256, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv63)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv64 = Conv2dUnit(i256, i512, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv64)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv65 = Conv2dUnit(i512, i256, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv65)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-1, ))
# self.conv66 = Conv2dUnit(i256, i512, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv66)
dd = 67
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv67 = torch.nn.Conv2d(i512, 3*(num_classes + 5), kernel_size=(1, 1))
aaaaaaa2(base_model.layers[i1], self.conv67)
dd = 68
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv68 = Conv2dUnit(i256, i128, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv68)
# self.upsample2 = torch.nn.Upsample(scale_factor=2, mode='nearest')
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv69 = Conv2dUnit(i128+i256, i128, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv69)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv70 = Conv2dUnit(i128, i256, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv70)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv71 = Conv2dUnit(i256, i128, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv71)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv72 = Conv2dUnit(i128, i256, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv72)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv73 = Conv2dUnit(i256, i128, (1, 1), stride=1, padding=0)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv73)
dd += 1
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd-2, ))
# self.conv74 = Conv2dUnit(i128, i256, (3, 3), stride=1, padding=1)
aaaaaaa(base_model.layers[i1], base_model.layers[i2], self.conv74)
dd = 75
i1, i2 = find(base_model, 'conv2d_%d'%(dd, ), 'batch_normalization_%d'%(dd, ))
# self.conv75 = torch.nn.Conv2d(i256, 3*(num_classes + 5), kernel_size=(1, 1))
aaaaaaa2(base_model.layers[i1], self.conv75)
base_model.save('qqq.h5')
# keras.utils.vis_utils.plot_model(base_model, to_file='aaa.png', show_shapes=True)