/
vgg16_class_AE.py
386 lines (321 loc) · 19.9 KB
/
vgg16_class_AE.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
"""
cifar10vgg.py
geifmany/cifar-vgg
https://github.com/geifmany/cifar-vgg/blob/master/cifar10vgg.py
"""
import numpy as np
import matplotlib.pyplot as plt
from scipy.misc import toimage
import keras
from keras.datasets import cifar10
from keras.models import Model
from keras.utils import to_categorical
from keras.models import Sequential
from keras import initializers, layers
from keras.layers import Input, Convolution2D, MaxPooling2D, UpSampling2D
from keras.layers import Dense, Dropout, Activation, Flatten
from keras import layers, models, optimizers
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import UpSampling2D, Conv2D
import os
from keras import regularizers
def save_history(history, result_file,epochs):
loss = history.history['loss']
#conv_loss = history.history['conv_out_loss']
acc = history.history['conv_out_acc']
cat_acc = history.history['softmax_acc']
val_loss = history.history['val_loss']
#val_conv_loss = history.history['val_out_recon_loss']
val_acc = history.history['val_conv_out_acc']
val_cat_acc = history.history['val_softmax_acc']
nb_epoch = len(acc)
with open(result_file, "a") as fp:
if epochs==0:
fp.write("i\tloss\tconv_acc\tcat_acc\tval_loss\tval_conv_acc\tval_cat_acc\n")
for i in range(nb_epoch):
fp.write("%d\t%f\t%f\t%f\t%f\t%f\t%f\n" % (epochs, loss[i], acc[i], cat_acc[i], val_loss[i], val_acc[i], val_cat_acc[i]))
else:
for i in range(nb_epoch):
fp.write("%d\t%f\t%f\t%f\t%f\t%f\t%f\n" % (epochs, loss[i], acc[i], cat_acc[i], val_loss[i], val_acc[i], val_cat_acc[i]))
if __name__ == '__main__':
# CIFAR-10データセットをロード
(X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
print(X_train.shape, Y_train.shape)
print(X_test.shape, Y_test.shape)
# 画像を描画
nclasses = 10
pos = 1
for targetClass in range(nclasses):
targetIdx = []
# クラスclassIDの画像のインデックスリストを取得
for i in range(len(Y_train)):
if Y_train[i][0] == targetClass:
targetIdx.append(i)
# 各クラスからランダムに選んだ最初の10個の画像を描画
np.random.shuffle(targetIdx)
for idx in targetIdx[:10]:
img = toimage(X_train[idx])
plt.subplot(10, 10, pos)
plt.imshow(img)
plt.axis('off')
pos += 1
plt.pause(3)
plt.close()
# 入力画像の次元
img_rows, img_cols = 32, 32
# チャネル数（RGBなので3）
img_channels = 3
# 画素値を0-1に変換
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
X_train /= 255.0
X_test /= 255.0
nb_classes = 10
# クラスラベル（0-9）をone-hotエンコーディング形式に変換
Y_train = to_categorical(Y_train, nb_classes)
Y_test = to_categorical(Y_test, nb_classes)
def vgg16_model(input_shape, n_class=10):
# Build the network of vgg for 10 classes with massive dropout and weight decay as described in the paper.
weight_decay = 0.0 #0.0005
axis_num = -1
x = layers.Input(shape=input_shape)
y = layers.Input(shape=(n_class,))
conv1 = Conv2D(64, (3, 3),activation='relu', padding='same',
input_shape=(32, 32, 3), kernel_regularizer=regularizers.l2(weight_decay))(x)
conv1 = BatchNormalization()(conv1)
conv1 = Dropout(0.3)(conv1) #0.3
conv2 = Conv2D(64, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv1)
conv2 = BatchNormalization()(conv2)
conv2 = Dropout(0.3)(conv2) #add 0.3
conv2 = MaxPooling2D(pool_size=(2, 2))(conv2) #>>>conv_out1
conv3 = Conv2D(128, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv2)
conv3 = BatchNormalization()(conv3)
conv3 = Dropout(0.4)(conv3) #0.4
conv4 = Conv2D(128, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv3)
conv4 = BatchNormalization()(conv4)
conv4 = MaxPooling2D(pool_size=(2, 2))(conv4)
conv5= Conv2D(256, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv4)
conv5 = BatchNormalization()(conv5)
conv5 = Dropout(0.4)(conv5) #0.4
conv6= Conv2D(256, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv5)
conv6 = BatchNormalization()(conv6)
conv6 = Dropout(0.4)(conv6) #0.4
conv7= Conv2D(256, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv6)
conv7 = BatchNormalization()(conv7)
conv7 = MaxPooling2D(pool_size=(2, 2))(conv7)
conv8= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv7)
conv8 = BatchNormalization()(conv8)
conv8 = Dropout(0.4)(conv8)
conv9= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv8)
conv9 = BatchNormalization()(conv9)
conv9 = Dropout(0.4)(conv9) #>>>conv_out2
conv10= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv9)
conv10 = BatchNormalization()(conv10)
conv10 = MaxPooling2D(pool_size=(2, 2))(conv10)
conv11= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv10)
conv11 = BatchNormalization()(conv11)
conv11 = Dropout(0.4)(conv11)
conv12= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv11)
conv12 = BatchNormalization()(conv12)
conv12 = Dropout(0.4)(conv12)
conv13= Conv2D(512, (3, 3),activation='relu', padding='same',kernel_regularizer=regularizers.l2(weight_decay))(conv12)
conv13 = BatchNormalization()(conv13)
conv13 = MaxPooling2D(pool_size=(2, 2))(conv13)
conv13 = Dropout(0.4)(conv13)
x1=Flatten()(conv13)
x1=Dense(512,activation='relu',kernel_regularizer=regularizers.l2(weight_decay))(x1)
x1 = BatchNormalization()(x1)
x1 = Dropout(0.5)(x1)
softmax=Dense(n_class, activation='softmax',name='softmax')(x1)
conv14=Conv2D(128, (3, 3),activation='relu', padding="same")(conv9)
conv14=UpSampling2D(size=(2, 2))(conv14)
conv14=Conv2D(128, (3, 3),activation='relu', padding="same")(conv14)
conv14=UpSampling2D(size=(2, 2))(conv14)
conv14=Conv2D(64, (3, 3),activation='relu', padding="same")(conv14)
conv14=UpSampling2D(size=(2, 2))(conv14)
"""
conv14=Conv2D(32, (3, 3),activation='relu', padding="same")(conv14)
conv14=UpSampling2D(size=(2, 2))(conv14)
conv14=Conv2D(32, (3, 3),activation='relu', padding="same")(conv14)
conv14=UpSampling2D(size=(2, 2))(conv14)
"""
conv15=UpSampling2D(size=(2, 2))(conv2)
conv_out=Conv2D(3, (3, 3), padding="same",activation='sigmoid', name="conv_out")(conv15)
#conv16=UpSampling2D(size=(2, 2))(conv14)
#conv_out=Conv2D(3, (3, 3), padding="same",activation='sigmoid', name="conv_out")(conv16)
return models.Model([x, y], [softmax, conv_out])
def train(model, data, epoch_size=32):
(x_train, y_train), (x_test, y_test) = data
#loss='binary_crossentropy' 'categorical_crossentropy' 'mse'
model.compile(optimizer='adam',
loss={'softmax':'categorical_crossentropy', 'conv_out':'mse'},
loss_weights=[1., 100.],
metrics={'softmax': 'accuracy','conv_out': 'accuracy'})
history=model.fit([x_train, y_train],[y_train, x_train], batch_size=32, epochs=epoch_size,
validation_data=[[x_test, y_test], [y_test, x_test]])
return model, history
def to3d(X):
if X.shape[-1]==3: return X
b = X.transpose(3,1,2,0)
c = np.array([b[0],b[0],b[0]])
return c.transpose(3,1,2,0)
def plot_generated_batch(i, model,data1,data2):
x_test, y_test = data1
y_pred, x_recon = model.predict([x_test, y_test], batch_size=32)
X_gen = x_recon
X_raw = x_test
Xs1 = to3d(X_raw[:10])
Xg1 = to3d(X_gen[:10])
Xs1 = np.concatenate(Xs1, axis=1)
Xg1 = np.concatenate(Xg1, axis=1)
Xs2 = to3d(X_raw[10:20])
Xg2 = to3d(X_gen[10:20])
Xs2 = np.concatenate(Xs2, axis=1)
Xg2 = np.concatenate(Xg2, axis=1)
x_train, y_train = data2
y_pred, x_recon = model.predict([x_train, y_train], batch_size=32)
X_gen = x_recon
X_raw = x_train
Xs3 = to3d(X_raw[:10])
Xg3 = to3d(X_gen[:10])
Xs3 = np.concatenate(Xs3, axis=1)
Xg3 = np.concatenate(Xg3, axis=1)
Xs4 = to3d(X_raw[10:20])
Xg4 = to3d(X_gen[10:20])
Xs4 = np.concatenate(Xs4, axis=1)
Xg4 = np.concatenate(Xg4, axis=1)
XX = np.concatenate((Xs1,Xg1,Xs2,Xg2,Xs3,Xg3,Xs4,Xg4), axis=0)
plt.imshow(XX)
plt.axis('off')
plt.savefig("./caps_figures/vggAutoCrassiEncoder{0:03d}.png".format(i))
plt.pause(3)
plt.close()
input_img = Input(shape=(32,32,3))
model = vgg16_model(input_shape=[32, 32, 3], n_class=10)
# モデルのサマリを表示
model.summary()
X_train = np.array(X_train)
Y_train = np.array(Y_train)
X_test = np.array(X_test)
Y_test = np.array(Y_test)
# 訓練
#model, history = train(model=model, data=((X_train, Y_train), (X_test, Y_test)), epoch_size=1)
for j in range(20):
model, history = train(model=model, data=((X_train, Y_train), (X_test, Y_test)), epoch_size=1)
model.save_weights('params_vggAE_epoch_{0:03d}.hdf5'.format(j), True)
plot_generated_batch(j,model=model, data1=(X_train, Y_train),data2=(X_test, Y_test))
# 学習履歴を保存
save_history(history, os.path.join("./caps_figures/", 'history_vgg_ACE.txt'),j)
"""
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_2 (InputLayer) (None, 32, 32, 3) 0
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 32, 32, 64) 1792 input_2[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 32, 32, 64) 256 conv2d_1[0][0]
__________________________________________________________________________________________________
dropout_1 (Dropout) (None, 32, 32, 64) 0 batch_normalization_1[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 32, 32, 64) 36928 dropout_1[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 32, 32, 64) 256 conv2d_2[0][0]
__________________________________________________________________________________________________
dropout_2 (Dropout) (None, 32, 32, 64) 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 16, 16, 64) 0 dropout_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 16, 16, 128) 73856 max_pooling2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 16, 16, 128) 512 conv2d_3[0][0]
__________________________________________________________________________________________________
dropout_3 (Dropout) (None, 16, 16, 128) 0 batch_normalization_3[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 16, 16, 128) 147584 dropout_3[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 16, 16, 128) 512 conv2d_4[0][0]
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, 8, 8, 128) 0 batch_normalization_4[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 8, 8, 256) 295168 max_pooling2d_2[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 8, 8, 256) 1024 conv2d_5[0][0]
__________________________________________________________________________________________________
dropout_4 (Dropout) (None, 8, 8, 256) 0 batch_normalization_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 8, 8, 256) 590080 dropout_4[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, 8, 8, 256) 1024 conv2d_6[0][0]
__________________________________________________________________________________________________
dropout_5 (Dropout) (None, 8, 8, 256) 0 batch_normalization_6[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 8, 8, 256) 590080 dropout_5[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, 8, 8, 256) 1024 conv2d_7[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, 4, 4, 256) 0 batch_normalization_7[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 4, 4, 512) 1180160 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, 4, 4, 512) 2048 conv2d_8[0][0]
__________________________________________________________________________________________________
dropout_6 (Dropout) (None, 4, 4, 512) 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 4, 4, 512) 2359808 dropout_6[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, 4, 4, 512) 2048 conv2d_9[0][0]
__________________________________________________________________________________________________
dropout_7 (Dropout) (None, 4, 4, 512) 0 batch_normalization_9[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 4, 4, 512) 2359808 dropout_7[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, 4, 4, 512) 2048 conv2d_10[0][0]
__________________________________________________________________________________________________
max_pooling2d_4 (MaxPooling2D) (None, 2, 2, 512) 0 batch_normalization_10[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 2, 2, 512) 2359808 max_pooling2d_4[0][0]
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None, 2, 2, 512) 2048 conv2d_11[0][0]
__________________________________________________________________________________________________
dropout_8 (Dropout) (None, 2, 2, 512) 0 batch_normalization_11[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, 2, 2, 512) 2359808 dropout_8[0][0]
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None, 2, 2, 512) 2048 conv2d_12[0][0]
__________________________________________________________________________________________________
dropout_9 (Dropout) (None, 2, 2, 512) 0 batch_normalization_12[0][0]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, 2, 2, 512) 2359808 dropout_9[0][0]
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None, 2, 2, 512) 2048 conv2d_13[0][0]
__________________________________________________________________________________________________
max_pooling2d_5 (MaxPooling2D) (None, 1, 1, 512) 0 batch_normalization_13[0][0]
__________________________________________________________________________________________________
dropout_10 (Dropout) (None, 1, 1, 512) 0 max_pooling2d_5[0][0]
__________________________________________________________________________________________________
flatten_1 (Flatten) (None, 512) 0 dropout_10[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 512) 262656 flatten_1[0][0]
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None, 512) 2048 dense_1[0][0]
__________________________________________________________________________________________________
dropout_11 (Dropout) (None, 512) 0 batch_normalization_14[0][0]
__________________________________________________________________________________________________
up_sampling2d_4 (UpSampling2D) (None, 32, 32, 64) 0 max_pooling2d_1[0][0]
__________________________________________________________________________________________________
softmax (Dense) (None, 10) 5130 dropout_11[0][0]
__________________________________________________________________________________________________
conv_out (Conv2D) (None, 32, 32, 3) 1731 up_sampling2d_4[0][0]
==================================================================================================
Total params: 15,003,149
Trainable params: 14,993,677
Non-trainable params: 9,472
__________________________________________________________________________________________________
Train on 50000 samples, validate on 10000 samples
j=0 Epoch 1/1
2018-04-02 06:05:24.458929: I C:\tf_jenkins\home\workspace\rel-win\M\windows-gpu\PY\35\tensorflow\core\common_runtime\gpu\gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GTX 1080, pci bus id: 0000:01:00.0, compute capability: 6.1)
50000/50000 [==============================] - 117s 2ms/step - loss: 2.6721 - softmax_loss: 1.9839 - conv_out_loss: 0.0069 - softmax_acc: 0.3062 - conv_out_acc: 0.6681 - val_loss: 1.9908 - val_softmax_loss: 1.5698 - val_conv_out_loss: 0.0042 - val_softmax_acc: 0.4342 - val_conv_out_acc: 0.7466
j=19 epoch 1/1
50000/50000 [==============================] - 120s 2ms/step - loss: 0.4092 - softmax_loss: 0.2581 - conv_out_loss: 0.0015 - softmax_acc: 0.9147 - conv_out_acc: 0.8163 - val_loss: 0.7814 - val_softmax_loss: 0.4931 - val_conv_out_loss: 0.0029 - val_softmax_acc: 0.8553 - val_conv_out_acc: 0.8482
"""