-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathldg_v2.py
156 lines (129 loc) · 6.49 KB
/
ldg_v2.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
from keras.layers import Lambda, Input, Dense, Flatten, Conv2D, Conv2DTranspose
from keras.layers import Activation, BatchNormalization, Reshape, Concatenate, Dropout
from keras.models import Model
from keras.losses import mse, binary_crossentropy
from keras.callbacks import Callback, ModelCheckpoint, ReduceLROnPlateau
from keras.utils import to_categorical
from keras import backend as K
from keras.optimizers import Adam
from keras import initializers
import numpy as np
import os
from mnist import MNIST
label_str = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
lebel_dict = {i: v for i, v in enumerate(label_str)}
n_label = len(label_str)
img_rows = 28
img_cols = 28
img_channel = 1
orig_dimension = img_rows * img_cols
image_shape = (img_rows, img_cols, img_channel)
batch_size = 128
epochs = 2
latent_dim = 10
learning_rate = 0.001
w_init = initializers.random_normal(stddev=0.02)
gamma_init = initializers.random_normal(mean=1.0, stddev=0.02)
def load_preprocess_EMNIST():
emndata = MNIST('emnist_data')
X_train, y_train = emndata.load('emnist_data/emnist-byclass-train-images-idx3-ubyte',
'emnist_data/emnist-byclass-train-labels-idx1-ubyte')
X_test, y_test = emndata.load('emnist_data/emnist-byclass-test-images-idx3-ubyte',
'emnist_data/emnist-byclass-test-labels-idx1-ubyte')
X_train = np.array(X_train)
y_train = np.array(y_train)
X_test = np.array(X_test)
y_test = np.array(y_test)
X_train = X_train.astype('float32') / (255 / 2) - 1 # for tanh
X_test = X_test.astype('float32') / (255 / 2) - 1
y_train = to_categorical(y_train, n_label)
y_test = to_categorical(y_test, n_label)
return X_train, y_train, X_test, y_test
def sampling(arg):
z_mean = arg[0]
z_log_var = arg[1]
# arg = [z_mean, z_log_var]
dim = K.int_shape(z_mean)[1]
# reparameterization trick
epsilon = K.random_normal(
shape=(K.shape(z_mean)[0], dim), mean=0.0, stddev=1.0)
return z_mean + K.exp(0.5 * z_log_var) * epsilon
def build_vae():
img_inputs = Input(shape=(orig_dimension,), name='image_input')
label_inputs = Input(shape=(n_label,), name='label_input')
encoder_inputs = Concatenate()([img_inputs, label_inputs])
x = Dense(orig_dimension, kernel_initializer=w_init,activation='relu')(encoder_inputs)
x = Reshape(image_shape)(x)
x = Conv2D(16, 3, strides=1, padding='same', kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
x = Conv2D(32, 3, strides=1, padding='same', kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
x = Conv2D(64, 3, strides=2, padding='same', kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
before_flatten_shape = K.int_shape(x)
x = Flatten()(x)
x = Dense(128, kernel_initializer=w_init, activation='relu')(x)
z_mean = Dense(latent_dim, name='z_mean')(x)
z_log_var = Dense(latent_dim, name='z_log_var')(x)
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])
encoder = Model([img_inputs, label_inputs], [z_mean, z_log_var, z], name='encoder')
latent_inputs = Input(shape=(latent_dim,), name='latent_inputs')
decoder_inputs = Concatenate()([latent_inputs, label_inputs])
x = Dense(128, kernel_initializer=w_init,
activation='relu')(decoder_inputs)
x = Dense(before_flatten_shape[1] * before_flatten_shape[2] * before_flatten_shape[3], activation='relu', kernel_initializer=w_init)(x)
x = Reshape((before_flatten_shape[1], before_flatten_shape[2], before_flatten_shape[3]))(x)
x = Conv2DTranspose(64, 3, strides=1, padding='same',kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
x = Conv2DTranspose(32, 3, strides=2, padding='same',kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
x = Conv2DTranspose(16, 3, strides=1, padding='same',kernel_initializer=w_init)(x)
x = BatchNormalization(gamma_initializer=gamma_init)(x)
x = Activation('relu')(x)
x = Conv2DTranspose(img_channel, 3, activation='tanh',padding='same', kernel_initializer=w_init)(x)
outputs = Flatten()(x)
decoder = Model([latent_inputs, label_inputs], outputs, name='decoder')
outputs = decoder([encoder([img_inputs, label_inputs])[2], label_inputs])
vae = Model([img_inputs, label_inputs], outputs)
beta = 1 # 1 --> regular VAE
reconstruction_loss = mse(img_inputs, outputs)
reconstruction_loss *= img_rows * img_cols
kl_loss = -0.5 * beta * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
vae_loss = K.mean(reconstruction_loss + kl_loss)
# in this case, we set y value to NONE during the fit
vae.add_loss(vae_loss)
vae.compile(optimizer=Adam(lr=learning_rate))
return vae
def train_or_load_weights(flag, weight_name=None):
saving_folder = 'best_weight_ldg_v2_conv-cvae'
if flag == 'train':
file_name = 'ldg_v2_conv-cvae'
os.makedirs(saving_folder, exist_ok=True)
best_model_weight_path = os.path.join(
saving_folder, file_name + '-best-wiehgts' + '-{epoch:03d}-{loss:.3f}-{val_loss:.3f}.h5')
save_best_model = ModelCheckpoint(best_model_weight_path, monitor='val_loss',
verbose=0, save_weights_only=True, save_best_only=True, mode='min')
learning_rate_reduction = ReduceLROnPlateau(
monitor='val_loss', patience=5, verbose=1, factor=0.5, min_lr=0.00001)
hist = vae.fit([X_train, y_train],
epochs=epochs,
batch_size=batch_size,
validation_data=([X_test, y_test], None),
verbose=1,
callbacks=[save_best_model, learning_rate_reduction]) # ,NEpochPrinter])
return hist
if flag == 'load':
# load all the weights for encoder and decoder when loading for vae
vae.load_weights(os.path.join(saving_folder, weight_name))
def load_weights_vae(weight_name=None):
# load all the weights for encoder and decoder when loading for vae
vae.load_weights(os.path.join(saving_folder, weight_name))
if __name__ == '__main__':
X_train, y_train, X_test, y_test = load_preprocess_EMNIST()
vae = build_vae()
train_or_load_weights(flag='train')