-
Notifications
You must be signed in to change notification settings - Fork 24
/
cifar10_generate.py
156 lines (132 loc) · 5.79 KB
/
cifar10_generate.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
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
import keras
from keras.layers import Input, Dense, Lambda, Flatten, Reshape, Layer
from keras.layers import Conv2D, Conv2DTranspose
from keras.models import Model
from keras import backend as K
from keras import metrics
from keras.datasets import cifar10
import cPickle
# import parameters
from cifar10_params import *
"""
loading vae model back is not a straight-forward task because of custom loss layer.
we have to define some architecture back again to specify custom loss layer and hence to load model back again.
"""
# tensorflow or theano
if K.image_data_format() == 'channels_first':
original_img_size = (img_chns, img_rows, img_cols)
else:
original_img_size = (img_rows, img_cols, img_chns)
# encoder architecture
x = Input(shape=original_img_size)
conv_1 = Conv2D(img_chns,
kernel_size=(2, 2),
padding='same', activation='relu')(x)
conv_2 = Conv2D(filters,
kernel_size=(2, 2),
padding='same', activation='relu',
strides=(2, 2))(conv_1)
conv_3 = Conv2D(filters,
kernel_size=num_conv,
padding='same', activation='relu',
strides=1)(conv_2)
conv_4 = Conv2D(filters,
kernel_size=num_conv,
padding='same', activation='relu',
strides=1)(conv_3)
flat = Flatten()(conv_4)
hidden = Dense(intermediate_dim, activation='relu')(flat)
z_mean = Dense(latent_dim)(hidden)
z_log_var = Dense(latent_dim)(hidden)
# Custom loss layer
class CustomVariationalLayer(Layer):
def __init__(self, **kwargs):
self.is_placeholder = True
super(CustomVariationalLayer, self).__init__(**kwargs)
def vae_loss(self, x, x_decoded_mean_squash):
x = K.flatten(x)
x_decoded_mean_squash = K.flatten(x_decoded_mean_squash)
xent_loss = img_rows * img_cols * metrics.binary_crossentropy(x, x_decoded_mean_squash)
kl_loss = - 0.5 * K.mean(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
return K.mean(xent_loss + kl_loss)
def call(self, inputs):
x = inputs[0]
x_decoded_mean_squash = inputs[1]
loss = self.vae_loss(x, x_decoded_mean_squash)
self.add_loss(loss, inputs=inputs)
# We don't use this output.
return x
# load saved models
vae = keras.models.load_model('../models/cifar10_ld_%d_conv_%d_id_%d_e_%d_vae.h5' % (latent_dim, num_conv, intermediate_dim, epochs),
custom_objects={'latent_dim':latent_dim, 'epsilon_std':epsilon_std, 'CustomVariationalLayer':CustomVariationalLayer})
encoder = keras.models.load_model('../models/cifar10_ld_%d_conv_%d_id_%d_e_%d_encoder.h5' % (latent_dim, num_conv, intermediate_dim, epochs),
custom_objects={'latent_dim':latent_dim, 'epsilon_std':epsilon_std, 'CustomVariationalLayer':CustomVariationalLayer})
generator = keras.models.load_model('../models/cifar10_ld_%d_conv_%d_id_%d_e_%d_generator.h5' % (latent_dim, num_conv, intermediate_dim, epochs),
custom_objects={'latent_dim':latent_dim, 'epsilon_std':epsilon_std, 'CustomVariationalLayer':CustomVariationalLayer})
# load history if saved
fname = '../models/cifar10_ld_%d_conv_%d_id_%d_e_%d_history.pkl' % (latent_dim, num_conv, intermediate_dim, epochs)
try:
with open(fname, 'rb') as fo:
history = cPickle.load(fo)
print history
except:
print "training history not saved"
# load dataset to plot latent space
(x_train, _), (x_test, y_test) = cifar10.load_data()
x_train = x_train.astype('float32') / 255.
x_train = x_train.reshape((x_train.shape[0],) + original_img_size)
x_test = x_test.astype('float32') / 255.
x_test = x_test.reshape((x_test.shape[0],) + original_img_size)
if latent_dim == 3:
x_test_encoded = encoder.predict(x_test, batch_size=batch_size)
fig = plt.figure(figsize=(12,12))
ax = fig.add_subplot(111, projection='3d')
ax.scatter(x_test_encoded[:, 0], x_test_encoded[:, 1],x_test_encoded[:, 2], c=y_test)
plt.show()
if latent_dim == 2:
# display a 2D plot of the classes in the latent space
x_test_encoded = encoder.predict(x_test, batch_size=batch_size)
plt.figure(figsize=(6, 6))
plt.scatter(x_test_encoded[:, 0], x_test_encoded[:, 1], c=y_test)
plt.colorbar()
plt.show()
"""
# display a 2D manifold of the images
n = 15 # figure with 15x15 images
img_size = 32
figure = np.zeros((img_size * n, img_size * n, img_chns))
# linearly spaced coordinates on the unit square were transformed through the inverse CDF (ppf) of the Gaussian
# to produce values of the latent variables z, since the prior of the latent space is Gaussian
grid_x = norm.ppf(np.linspace(0.05, 0.95, n))
grid_y = norm.ppf(np.linspace(0.05, 0.95, n))
for i, yi in enumerate(grid_x):
for j, xi in enumerate(grid_y):
z_sample = np.array([[xi, yi]])
z_sample = np.tile(z_sample, batch_size).reshape(batch_size, 2)
x_decoded = generator.predict(z_sample, batch_size=batch_size)
img = x_decoded[0].reshape(img_size, img_size, img_chns)
figure[i * img_size: (i + 1) * img_size,
j * img_size: (j + 1) * img_size] = img
plt.figure(figsize=(10, 10))
plt.imshow(figure, cmap='Greys_r')
plt.show()
"""
# display images generated from randomly sampled latent vector
n = 15
img_size = 32
figure = np.zeros((img_size * n, img_size * n, img_chns))
for i in range(n):
for j in range(n):
z_sample = np.array([np.random.uniform(-1,1 ,size=latent_dim)])
x_decoded = generator.predict(z_sample)
img = x_decoded[0].reshape(img_size, img_size, img_chns)
figure[i * img_size: (i + 1) * img_size,j * img_size: (j + 1) * img_size] = img
#plt.figure(figsize=(5, 5))
#plt.imshow(img, cmap='Greys_r')
#plt.show()
plt.figure(figsize=(20, 20))
plt.imshow(figure, cmap='Greys_r')
plt.show()