-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathEvANet_keras.py
163 lines (139 loc) · 6.34 KB
/
EvANet_keras.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
from keras.datasets import cifar10
from keras.layers.convolutional import Conv2D
from keras.layers.normalization import BatchNormalization
from keras.layers.core import Activation, Dense, Flatten
from keras.layers import Input
from keras.layers.merge import Add, Concatenate
from keras.layers.pooling import MaxPooling2D, AveragePooling2D
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Model
from keras.utils import to_categorical
def conv_norm_block(in_blob, width, filter_size):
tmpnet = Conv2D(width, filter_size, padding='same')(in_blob)
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Activation('relu')(tmpnet)
return tmpnet
def res_block_2015(in_blob, width, filter_size):
tmpnet = conv_norm_block(in_blob, width, filter_size)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
tmpnet = Activation('relu')(tmpnet)
return tmpnet
def res_block_2016(in_blob, width, filter_size):
tmpnet = BatchNormalization()(in_blob)
tmpnet = Activation('relu')(tmpnet)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Activation('relu')(tmpnet)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
return tmpnet
def no_relu(in_blob, width, filter_size):
tmpnet = conv_norm_block(in_blob, width, filter_size)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
return tmpnet
def res_conv_block_2016(in_blob, width, filter_size):
tmpnet = BatchNormalization()(in_blob)
tmpnet = Activation('relu')(tmpnet)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Activation('relu')(tmpnet)
tmpnet = Conv2D(widht, filter_size, padding='same')(tmpnet)
shortcut = BatchNormalization()(in_blob)
shortcut = Activation('relu')(shortcut)
shortcut = Conv2D(width, filter_size, padding='same')(shortcut)
tmpnet = Add()([shortcut, tmpnet])
return tmpnet
def no_act_block_2015(in_blob, widht, filter_size):
tmpnet = conv_norm_block(in_blob, width, filter_size)
tmpnet = Conv2D(widht, filter_size, padding='same')(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
return tmpnet
def bn_after_add_block(in_blob, width, filter_size):
tmpnet = conv_norm_block(in_blob, width, filter_size)
tmpnet = Conv2D(width, filter_size, padding='same')(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
tmpnet = BatchNormalization()(tmpnet)
tmpnet = Activation('relu')(tmpnet)
return tmpnet
def inception_v1_block(in_blob, inc1_width, inc3_width, inc5_width, pool_width,
out_width):
tmpnet = BatchNormalization()(in_blob)
tmpnet = Activation('relu')(tmpnet)
inc1net = Conv2D(inc1_width, 1, padding='same')(tmpnet)
inc1net = Activation('relu')(inc1net)
inc3net = Conv2D(inc3_width//2, 1, padding='same')(tmpnet)
inc3net = Activation('relu')(inc3net)
inc3net = Conv2D(inc3_width, 3, padding='same')(inc3net)
inc3net = Activation('relu')(inc3net)
inc5net = Conv2D(inc5_width//2, 1, padding='same')(tmpnet)
inc5net = Activation('relu')(inc5net)
inc5net = Conv2D(inc5_width, 5, padding='same')(inc5net)
inc5net = Activation('relu')(inc5net)
poolnet = MaxPooling2D(pool_size=(3,3), strides=(1,1), padding='same')(tmpnet)
poolnet = Conv2D(pool_width, 1, padding='same')(poolnet)
poolnet = Activation('relu')(poolnet)
tmpnet = Concatenate(axis=3)([inc1net, inc3net, inc5net, poolnet])
tmpnet = Conv2D(out_width, 1, padding='same')(tmpnet)
tmpnet = Add()([in_blob, tmpnet])
return tmpnet
num_classes = 10
num_epochs = 200
batch_size = 128
(X, Y), (test_x, test_y) = cifar10.load_data()
#X = X.reshape([-1,32,32,3])
#test_x = test_x.reshape([-1,32,32,3])
X = X.astype('float32')
text_x = test_x.astype('float32')
#X /= 255
#test_x /= 255
Y = to_categorical(Y, num_classes)
test_y = to_categorical(test_y, num_classes)
datagen = ImageDataGenerator(featurewise_center=True,
featurewise_std_normalization = True,
width_shift_range=0.25, height_shift_range=0.25,
vertical_flip=True)
datagen.fit(X)
a = Input(shape=(32,32,3))
evanet = conv_norm_block(a, 16, 3)
evanet = inception_v1_block(evanet, 8, 16, 32, 8, 16)
evanet = inception_v1_block(evanet, 8, 16, 32, 8, 16)
evanet = res_block_2016(evanet, 16, 3)
evanet = inception_v1_block(evanet, 8, 16, 8, 32, 16)
evanet = inception_v1_block(evanet, 8, 16, 32, 8, 16)
evanet = inception_v1_block(evanet, 8, 16, 32, 8, 16)
evanet = inception_v1_block(evanet, 8, 16, 32, 8, 16)
evanet = Conv2D(32, 1, strides=2, padding='same')(evanet)
evanet = res_block_2015(evanet, 32, 3)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = inception_v1_block(evanet, 16, 32, 64, 16, 32)
evanet = Conv2D(64, 1, strides=2, padding='same')(evanet)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = inception_v1_block(evanet, 32, 64, 128, 32, 64)
evanet = BatchNormalization()(evanet)
evanet = Activation('relu')(evanet)
evanet = AveragePooling2D(pool_size=(3, 3), strides=(1,1),
padding='same')(evanet)
evanet = Flatten()(evanet)
evanet_out = Dense(num_classes, activation='softmax')(evanet)
model = Model(inputs=a, outputs=evanet_out)
model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['acc'])
print("X.shape: ", X.shape)
print("Y.shape: ", Y.shape)
print("test_x.shape: ", test_x.shape)
print("test_y.shape: ", test_y.shape)
model.fit_generator(datagen.flow(X,Y, batch_size=batch_size),
steps_per_epoch=(len(X)//batch_size), epochs=num_epochs)
#validation_data=(test_x, test_y))