forked from bojone/vae
-
Notifications
You must be signed in to change notification settings - Fork 0
/
vae_keras_celeba.py
149 lines (124 loc) · 4.34 KB
/
vae_keras_celeba.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
#! -*- coding: utf-8 -*-
import numpy as np
from scipy import misc
import glob
import imageio
from keras.models import Model
from keras.layers import *
from keras import backend as K
from keras.optimizers import Adam
from keras.callbacks import Callback
imgs = glob.glob('img_align_celeba/*.jpg')
np.random.shuffle(imgs)
height,width = misc.imread(imgs[0]).shape[:2]
center_height = int((height - width) / 2)
img_dim = 64
z_dim = 512
def imread(f):
x = misc.imread(f)
x = x[center_height:center_height+width, :]
x = misc.imresize(x, (img_dim, img_dim))
return x.astype(np.float32) / 255 * 2 - 1
def data_generator(batch_size=32):
X = []
while True:
np.random.shuffle(imgs)
for f in imgs:
X.append(imread(f))
if len(X) == batch_size:
X = np.array(X)
yield X,None
X = []
x_in = Input(shape=(img_dim, img_dim, 3))
x = x_in
x = Conv2D(z_dim/16, kernel_size=(5,5), strides=(2,2), padding='SAME')(x)
x = BatchNormalization()(x)
x = LeakyReLU(0.2)(x)
x = Conv2D(z_dim/8, kernel_size=(5,5), strides=(2,2), padding='SAME')(x)
x = BatchNormalization()(x)
x = LeakyReLU(0.2)(x)
x = Conv2D(z_dim/4, kernel_size=(5,5), strides=(2,2), padding='SAME')(x)
x = BatchNormalization()(x)
x = LeakyReLU(0.2)(x)
x = Conv2D(z_dim/2, kernel_size=(5,5), strides=(2,2), padding='SAME')(x)
x = BatchNormalization()(x)
x = LeakyReLU(0.2)(x)
x = Conv2D(z_dim, kernel_size=(5,5), strides=(2,2), padding='SAME')(x)
x = BatchNormalization()(x)
x = LeakyReLU(0.2)(x)
x = GlobalAveragePooling2D()(x)
encoder = Model(x_in, x)
encoder.summary()
map_size = K.int_shape(encoder.layers[-2].output)[1:-1]
# 解码层，也就是生成器部分
z_in = Input(shape=K.int_shape(x)[1:])
z = z_in
z = Dense(np.prod(map_size)*z_dim)(z)
z = Reshape(map_size + (z_dim,))(z)
z = Conv2DTranspose(z_dim/2, kernel_size=(5,5), strides=(2,2), padding='SAME')(z)
z = BatchNormalization()(z)
z = Activation('relu')(z)
z = Conv2DTranspose(z_dim/4, kernel_size=(5,5), strides=(2,2), padding='SAME')(z)
z = BatchNormalization()(z)
z = Activation('relu')(z)
z = Conv2DTranspose(z_dim/8, kernel_size=(5,5), strides=(2,2), padding='SAME')(z)
z = BatchNormalization()(z)
z = Activation('relu')(z)
z = Conv2DTranspose(z_dim/16, kernel_size=(5,5), strides=(2,2), padding='SAME')(z)
z = BatchNormalization()(z)
z = Activation('relu')(z)
z = Conv2DTranspose(3, kernel_size=(5,5), strides=(2,2), padding='SAME')(z)
z = Activation('tanh')(z)
decoder = Model(z_in, z)
decoder.summary()
class ScaleShift(Layer):
def __init__(self, **kwargs):
super(ScaleShift, self).__init__(**kwargs)
def call(self, inputs):
z, shift, log_scale = inputs
z = K.exp(log_scale) * z + shift
logdet = -K.sum(K.mean(log_scale, 0))
self.add_loss(logdet)
return z
z_shift = Dense(z_dim)(x)
z_log_scale = Dense(z_dim)(x)
u = Lambda(lambda z: K.random_normal(shape=K.shape(z)))(z_shift)
z = ScaleShift()([u, z_shift, z_log_scale])
x_recon = decoder(z)
x_out = Subtract()([x_in, x_recon])
recon_loss = 0.5 * K.sum(K.mean(x_out**2, 0)) + 0.5 * np.log(2*np.pi) * np.prod(K.int_shape(x_out)[1:])
z_loss = 0.5 * K.sum(K.mean(z**2, 0)) - 0.5 * K.sum(K.mean(u**2, 0))
vae_loss = recon_loss + z_loss
vae = Model(x_in, x_out)
vae.add_loss(vae_loss)
vae.compile(optimizer=Adam(1e-4))
def sample(path):
n = 9
figure = np.zeros((img_dim*n, img_dim*n, 3))
for i in range(n):
for j in range(n):
x_recon = decoder.predict(np.random.randn(1, *K.int_shape(x)[1:]))
digit = x_recon[0]
figure[i*img_dim: (i+1)*img_dim,
j*img_dim: (j+1)*img_dim] = digit
figure = (figure + 1) / 2 * 255
imageio.imwrite(path, figure)
class Evaluate(Callback):
def __init__(self):
import os
self.lowest = 1e10
self.losses = []
if not os.path.exists('samples'):
os.mkdir('samples')
def on_epoch_end(self, epoch, logs=None):
path = 'samples/test_%s.png' % epoch
sample(path)
self.losses.append((epoch, logs['loss']))
if logs['loss'] <= self.lowest:
self.lowest = logs['loss']
encoder.save_weights('./best_encoder.weights')
evaluator = Evaluate()
vae.fit_generator(data_generator(),
epochs=1000,
steps_per_epoch=1000,
callbacks=[evaluator])