-
Notifications
You must be signed in to change notification settings - Fork 0
/
vae.py
173 lines (133 loc) · 5.43 KB
/
vae.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
from __future__ import division, print_function, absolute_import
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
import tensorflow as tf
import os
from scipy import misc
import cv2
# from tensorflow.examples.tutorials.mnist import input_data
# mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)
learning_rate = 0.001
num_steps = 2000
batch_size = 64
image_dim = 784
hidden_dim = 521
latent_dim = 2
###########################################
# Obtain images paths
path = "frames/"
imagepaths = list()
walk = os.walk(path).__next__()
for sample in walk[2]:
if sample.endswith(".png"):
imagepaths.append(os.path.join(path, sample))
images = []
for img in imagepaths:
i = misc.imread(img, flatten=True).flatten()
i = i/255
images.append(i)
# images = np.asarray(images)
# dataset = tf.data.Dataset.from_tensor_slices(images).batch(batch_size).repeat()
def create_batches():
while (True):
for i in range(0,len(images),batch_size):
yield(images[i:i+batch_size])
iter = create_batches()
########################################################
def glo_rot_init(shape):
return tf.random_normal(shape=shape, stddev=1. / tf.sqrt(shape[0]/2.))
weights = {
'encoder_h1': tf.Variable(glo_rot_init([image_dim, hidden_dim])),
'z_mean': tf.Variable(glo_rot_init([hidden_dim, latent_dim])),
'z_std': tf.Variable(glo_rot_init([hidden_dim, latent_dim])),
'decoder_h1': tf.Variable(glo_rot_init([latent_dim, hidden_dim])),
'decoder_out': tf.Variable(glo_rot_init([hidden_dim, image_dim]))
}
biases = {
'encoder_b1': tf.Variable(glo_rot_init([hidden_dim])),
'z_mean': tf.Variable(glo_rot_init([latent_dim])),
'z_std': tf.Variable(glo_rot_init([latent_dim])),
'decoder_b1': tf.Variable(glo_rot_init([hidden_dim])),
'decoder_out': tf.Variable(glo_rot_init([image_dim]))
}
input_image = tf.placeholder(tf.float32, shape=[None, image_dim])
encoder = tf.matmul(input_image, weights['encoder_h1']) + biases['encoder_b1']
encoder = tf.nn.tanh(encoder)
z_mean = tf.matmul(encoder, weights['z_mean']) + biases['z_mean']
z_std = tf.matmul(encoder, weights['z_std']) + biases['z_std']
eps = tf.random_normal(tf.shape(z_std), dtype=tf.float32, mean=0., stddev=1.0, name='epsilon')
z = z_mean + tf.exp(z_std/2) * eps
decoder = tf.matmul(z, weights['decoder_h1']) + biases['decoder_b1']
decoder = tf.nn.tanh(decoder)
decoder = tf.matmul(decoder, weights['decoder_out']) + biases['decoder_out']
decoder = tf.nn.sigmoid(decoder)
def vae_loss(x_reconstructed, x_true):
encode_decode_loss = x_true * tf.log(x_reconstructed) + (1 - x_true) * tf.log(1 - x_reconstructed)
encode_decode_loss = -tf.reduce_sum(encode_decode_loss, 1)
kl_div_loss = 1 + z_std - tf.square(z_mean) - tf.exp(z_std)
kl_div_loss = -0.5 * tf.reduce_sum(kl_div_loss, 1)
return tf.reduce_mean(encode_decode_loss + kl_div_loss)
loss_op = vae_loss(decoder, input_image)
optimizer = tf.train.RMSPropOptimizer(learning_rate=learning_rate, momentum=0.1)
train_op = optimizer.minimize(loss_op)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
for i in range(1, num_steps+1):
batch_x = iter.__next__()
# batch_x, _ = mnist.train.next_batch(batch_size)
feed_dict = {input_image: batch_x}
_, l = sess.run([train_op, loss_op], feed_dict=feed_dict)
if i % 1000 == 0 or i == 1:
print('Step %i, Loss %f' % (i, l))
##########################
noise_input = tf.placeholder(tf.float32, shape=[None, latent_dim])
decoder = tf.matmul(noise_input, weights['decoder_h1']) + biases['decoder_b1']
decoder = tf.nn.tanh(decoder)
decoder = tf.matmul(decoder, weights['decoder_out']) + biases['decoder_out']
decoder = tf.nn.sigmoid(decoder)
n = 20
x_axis = np.linspace(-3, 3, n)
y_axis = np.linspace(-3, 3, n)
canvas = np.empty((28 *n, 28 * n))
for i, yi in enumerate(x_axis):
for j, xi in enumerate(y_axis):
z_mu = np.array([[xi, yi]] * batch_size)
x_mean = sess.run(decoder, feed_dict={noise_input: z_mu})
canvas[(n - i - 1) * 28:(n - i) * 28, j * 28:(j + 1) * 28] = x_mean[0].reshape(28, 28)
# plt.figure(figsize=(8, 10))
# Xi, Yi = np.meshgrid(x_axis, y_axis)
plt.imshow(canvas, origin="upper", cmap="gray")
plt.show()
##################
canvas = np.empty((28*5, 28*2))
input_image = tf.placeholder(tf.float32, shape=[None, image_dim])
encoder = tf.matmul(input_image, weights['encoder_h1']) + biases['encoder_b1']
encoder = tf.nn.tanh(encoder)
z_mean = tf.matmul(encoder, weights['z_mean']) + biases['z_mean']
z_std = tf.matmul(encoder, weights['z_std']) + biases['z_std']
eps = tf.random_normal(tf.shape(z_std), dtype=tf.float32, mean=0., stddev=1.0, name='epsilon')
z = z_mean + tf.exp(z_std/2) * eps
decoder = tf.matmul(z, weights['decoder_h1']) + biases['decoder_b1']
decoder = tf.nn.tanh(decoder)
decoder = tf.matmul(decoder, weights['decoder_out']) + biases['decoder_out']
decoder = tf.nn.sigmoid(decoder)
path = "vr_on/"
imagepaths = list()
walk = os.walk(path).__next__()
for sample in walk[2]:
if sample.endswith(".png"):
imagepaths.append(os.path.join(path, sample))
images = []
canvas = np.zeros(28*2)
for img in imagepaths:
i = misc.imread(img, flatten=True).flatten().reshape(1, image_dim)
i = i/255
imgcp = i
canvas
o = sess.run(decoder, feed_dict={input_image: i})
out = o.reshape(28, 28)
canvas = np.vstack((canvas, np.hstack((i.reshape(28, 28), out))))
plt.imshow(canvas, origin="upper", cmap="gray")
plt.show()