-
Notifications
You must be signed in to change notification settings - Fork 6
/
CVAE.py
352 lines (287 loc) · 17.6 KB
/
CVAE.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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
import os
import h5py
import re
import numpy as np
import tensorflow as tf
import cv2
from tensorflow.python.training import moving_averages
from tensorflow.python.ops import control_flow_ops
import matplotlib.pyplot as plt
import pdb
from utilities import guided_filter, sample_normal
################ batch normalization setting ################
MOVING_AVERAGE_DECAY = 0.9997
BN_EPSILON = 0.001
BN_DECAY = MOVING_AVERAGE_DECAY
UPDATE_OPS_COLLECTION = 'Derain_update_ops'
DERAIN_VARIABLES = 'Derain_variables'
n_latent = 256
#############################################################
num_feature = 16
KernelSize = 3
os.environ['CUDA_VISIBLE_DEVICES'] = "5" # select GPU device
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('num_h5_file', 2000,
"""number of training h5 files.""")
tf.app.flags.DEFINE_integer('num_patches', 500,
"""number of patches in each h5 file.""")
tf.app.flags.DEFINE_float('learning_rate', 0.01,
"""learning rate.""")
tf.app.flags.DEFINE_float('beta', 1e-4,
"""beta.""")
tf.app.flags.DEFINE_integer('epoch', 3,
"""epoch.""")
tf.app.flags.DEFINE_integer('batch_size', 32,
"""Batch size.""")
tf.app.flags.DEFINE_integer('num_channels', 3,
"""Number of the input's channels.""")
tf.app.flags.DEFINE_float('beta', 1e-4,
"""epoch.""")
tf.app.flags.DEFINE_integer('image_size', 80,
"""Size of the images.""")
tf.app.flags.DEFINE_integer('label_size', 80,
"""Size of the labels.""")
tf.app.flags.DEFINE_integer('num_samples', 10,
"""number of the samples.""")
tf.app.flags.DEFINE_string("data_path", "../fu_h5data/", "The path of h5 files")
tf.app.flags.DEFINE_string("save_model_path", "./cvae_model/", "The path of saving model")
# read h5 files
def read_data(file):
with h5py.File(file, 'r') as hf:
data = hf.get('data')
label = hf.get('label')
return np.array(data), np.array(label)
def vae_prior(noise, is_training):
regularizer = tf.contrib.layers.l2_regularizer(scale=1e-10)
initializer = tf.contrib.layers.xavier_initializer()
with tf.variable_scope("prior", reuse=None):
# layer 1
with tf.variable_scope('layer_1'):
output = tf.layers.conv2d(noise, filters=16, kernel_size=3, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_1')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_1')
output_shortcut = tf.nn.leaky_relu(output, name='relu_1')
for i in range(3):
with tf.variable_scope('layer_%d' % (i * 2 + 2)):
output = tf.layers.conv2d(output_shortcut, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 2)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 2)))
output = tf.nn.leaky_relu(output, name=('relu_%d' % (i * 2 + 2)))
with tf.variable_scope('layer_%d' % (i * 2 + 3)):
output = tf.layers.conv2d(output, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 3)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 3)))
output = tf.nn.leaky_relu(output, name=('relu_%d' % (i * 2 + 3)))
output_shortcut = tf.add(output_shortcut, output) # shortcut
with tf.variable_scope('layer_final'):
output = tf.layers.conv2d(output_shortcut, 1, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_26')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_26')
mn = tf.layers.conv2d(output, filters=1, kernel_size=3, strides=1, padding="SAME", activation=None)
sd = tf.layers.conv2d(output, filters=1, kernel_size=3, strides=1, padding="SAME", activation=None)
z = sample_normal(mn, sd, FLAGS.num_samples)
return z, mn, sd
def vae_encoder(X, noise, is_training):
regularizer = tf.contrib.layers.l2_regularizer(scale=1e-10)
initializer = tf.contrib.layers.xavier_initializer()
x_concat = tf.concat([X, noise], axis=3)
with tf.variable_scope("encoder", reuse=None):
with tf.variable_scope('layer_1'):
output = tf.layers.conv2d(x_concat, filters=16, kernel_size=3, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_1')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_1')
output_shortcut = tf.nn.leaky_relu(output, name='relu_1')
for i in range(3):
with tf.variable_scope('layer_%d' % (i * 2 + 2)):
output = tf.layers.conv2d(output_shortcut, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 2)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 2)))
output = tf.nn.leaky_relu(output, name=('relu_%d' % (i * 2 + 2)))
with tf.variable_scope('layer_%d' % (i * 2 + 3)):
output = tf.layers.conv2d(output, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 3)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 3)))
output = tf.nn.leaky_relu(output, name=('relu_%d' % (i * 2 + 3)))
output_shortcut = tf.add(output_shortcut, output) # shortcut
with tf.variable_scope('layer_final'):
output = tf.layers.conv2d(output_shortcut, 1, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_26')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_26')
mn = tf.layers.conv2d(output, filters=1, kernel_size=3, strides=1, padding="SAME", activation=None)
sd = tf.layers.conv2d(output, filters=1, kernel_size=3, strides=1, padding="SAME", activation=None)
z = sample_normal(mn, sd, FLAGS.num_samples)
return z, mn, sd
def vae_decoder(z, noise, is_training):
regularizer = tf.contrib.layers.l2_regularizer(scale=1e-10)
initializer = tf.contrib.layers.xavier_initializer()
noise = tf.tile(tf.expand_dims(noise, axis=0), [FLAGS.num_samples, 1, 1, 1, 1])
x_concat = tf.concat([z, noise], axis=-1)
output_list = []
with tf.variable_scope("decoder", reuse=None):
for n in range(FLAGS.num_samples):
with tf.variable_scope('layer_1', reuse=tf.AUTO_REUSE):
output = tf.layers.conv2d_transpose(x_concat[n], filters=16, kernel_size=3, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_1')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_1')
output_shortcut = tf.nn.relu(output, name='relu_1')
for i in range(3):
with tf.variable_scope('layer_%d' % (i * 2 + 2), reuse=tf.AUTO_REUSE):
output = tf.layers.conv2d_transpose(output_shortcut, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 2)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 2)))
output = tf.nn.relu(output, name=('relu_%d' % (i * 2 + 2)))
with tf.variable_scope('layer_%d' % (i * 2 + 3), reuse=tf.AUTO_REUSE):
output = tf.layers.conv2d_transpose(output, num_feature, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name=('conv_%d' % (i * 2 + 3)))
output = tf.layers.batch_normalization(output, training=is_training, name=('bn_%d' % (i * 2 + 3)))
output = tf.nn.leaky_relu(output, name=('relu_%d' % (i * 2 + 3)))
output_shortcut = tf.add(output_shortcut, output) # shortcut
with tf.variable_scope('layer_final', reuse=tf.AUTO_REUSE):
output = tf.layers.conv2d_transpose(output_shortcut, 1, KernelSize, padding='same',
kernel_initializer=initializer,
kernel_regularizer=regularizer, name='conv_26')
output = tf.layers.batch_normalization(output, training=is_training, name='bn_26')
img = tf.nn.relu(output + noise[n])
output_list.append(tf.expand_dims(img, axis=0))
all_img = tf.concat(output_list, axis=0)
img = tf.reduce_mean(all_img, axis=0)
return img, all_img
if __name__ == '__main__':
if not os.path.exists('single_results/'):
os.makedirs('single_results/')
images = tf.placeholder(tf.float32, shape=(None, FLAGS.image_size, FLAGS.image_size, FLAGS.num_channels)) # data
details = tf.placeholder(tf.float32, shape=(None, FLAGS.image_size, FLAGS.image_size, FLAGS.num_channels)) # label
labels = tf.placeholder(tf.float32,
shape=(None, FLAGS.label_size, FLAGS.label_size, FLAGS.num_channels)) # detail layer
is_training = tf.placeholder(tf.bool)
labels_r = labels[:, :, :, :1]
labels_g = labels[:, :, :, 1:2]
labels_b = labels[:, :, :, 2:3]
images_r = images[:, :, :, :1]
images_g = images[:, :, :, 1:2]
images_b = images[:, :, :, 2:3]
details_r = details[:, :, :, :1]
details_g = details[:, :, :, 1:2]
details_b = details[:, :, :, 2:3]
with tf.variable_scope('r_out'):
r_sample, r_mu, r_var = vae_encoder(labels_r, images_r, is_training)
pr_sample, pr_mu, pr_var = vae_prior(images_r, is_training)
r_out, all_r = vae_decoder(r_sample, images_r, is_training)
with tf.variable_scope('g_out'):
g_sample, g_mu, g_var = vae_encoder(labels_g, images_g, is_training)
pg_sample, pg_mu, pg_var = vae_prior(images_g, is_training)
g_out, all_g = vae_decoder(g_sample, images_g, is_training)
with tf.variable_scope('b_out'):
b_sample, b_mu, b_var = vae_encoder(labels_b, images_b, is_training)
pb_sample, pb_mu, pb_var = vae_prior(images_b, is_training)
b_out, all_b = vae_decoder(b_sample, images_b, is_training)
with tf.variable_scope('final_out'):
outputs = tf.concat([r_out, g_out, b_out], axis=-1)
r_kl_loss = 0.5 * tf.reduce_sum((tf.exp(r_var) + (r_mu - pr_mu) ** 2) / tf.exp(pr_var) - 1. + (pr_var - r_var),
axis=[1, 2, 3])
g_kl_loss = 0.5 * tf.reduce_sum((tf.exp(g_var) + (g_mu - pg_mu) ** 2) / tf.exp(pg_var) - 1. + (pg_var - g_var),
axis=[1, 2, 3])
b_kl_loss = 0.5 * tf.reduce_sum((tf.exp(b_var) + (b_mu - pb_mu) ** 2) / tf.exp(pb_var) - 1. + (pb_var - b_var),
axis=[1, 2, 3])
kl_loss = tf.reduce_mean(r_kl_loss+g_kl_loss+b_kl_loss)
r_recon_loss = tf.reduce_sum(tf.squared_difference(r_out, labels_r), axis=[1, 2, 3])
g_recon_loss = tf.reduce_sum(tf.squared_difference(g_out, labels_g), axis=[1, 2, 3])
b_recon_loss = tf.reduce_sum(tf.squared_difference(b_out, labels_b), axis=[1, 2, 3])
recon_loss = tf.reduce_mean(r_recon_loss+g_recon_loss+b_recon_loss) # MSE loss
loss = recon_loss + FLAGS.beta * kl_loss
lr_ = FLAGS.learning_rate # learning rate
lr = tf.placeholder(tf.float32, shape=[])
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
train_op = tf.train.AdamOptimizer(lr).minimize(loss)
var_list = tf.trainable_variables()
g_list = tf.global_variables()
bn_moving_vars = [g for g in g_list if 'moving_mean' in g.name]
bn_moving_vars += [g for g in g_list if 'moving_variance' in g.name]
var_list += bn_moving_vars
saver = tf.train.Saver(max_to_keep=5)
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.8 # GPU setting
config.gpu_options.allow_growth = True
data_path = FLAGS.data_path
save_path = FLAGS.save_model_path
epoch = int(FLAGS.epoch)
with tf.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
validation_data_name = "validation.h5"
validation_data, validation_label = read_data(data_path + validation_data_name) # data for validation
validation_detail = validation_data - guided_filter(validation_data) # detail layer for validation
if tf.train.get_checkpoint_state('./cvae_model/'): # load previous trained models
ckpt = tf.train.latest_checkpoint('./cvae_model/')
saver.restore(sess, ckpt)
ckpt_num = re.findall(r"\d", ckpt)
if len(ckpt_num) == 2:
start_point = 10 * int(ckpt_num[0]) + int(ckpt_num[1])
else:
start_point = int(ckpt_num[0])
print("Load success")
else: # re-training if no previous trained models
print("re-training")
start_point = 0
for j in range(start_point, epoch): # the number of epoch
if j + 1 > 1: # reduce learning rate
lr_ = FLAGS.learning_rate * 0.1
if j + 1 > 2:
lr_ = FLAGS.learning_rate * 0.01
Training_Loss = 0.
for h5_num in range(FLAGS.num_h5_file): # the number of h5 files
train_data_name = "train" + str(h5_num + 1) + ".h5"
train_data, train_label = read_data(data_path + train_data_name) # data for training
detail_data = train_data - guided_filter(train_data) # detail layer for training
data_size = int(FLAGS.num_patches / FLAGS.batch_size) # the number of batch
train_loss = 0.0
train_r_loss = 0.0
train_g_loss = 0.0
train_b_loss = 0.0
train_rk_loss = 0.0
train_gk_loss = 0.0
train_bk_loss = 0.0
for batch_num in range(data_size):
rand_index = np.arange(int(batch_num * FLAGS.batch_size), int((batch_num + 1) * FLAGS.batch_size))
batch_data = train_data[rand_index, :, :, :]
batch_detail = detail_data[rand_index, :, :, :]
batch_label = train_label[rand_index, :, :, :]
_, lossvalue, rk_loss, gk_loss, bk_loss, r_loss, g_loss, b_loss = sess.run(
[train_op, loss, r_kl_loss, g_kl_loss, b_kl_loss, r_recon_loss, g_recon_loss, b_recon_loss],
feed_dict={images: batch_data,
labels: batch_label, lr: lr_,
is_training: True})
Training_Loss += lossvalue # training loss
train_loss += lossvalue
train_r_loss += r_loss
train_g_loss += g_loss
train_b_loss += b_loss
train_rk_loss += rk_loss
train_gk_loss += gk_loss
train_bk_loss += bk_loss
print('training %d epoch, %d / %d h5 files are finished, learning rate = %.4f, Training_Loss = %.4f' %
(j + 1, h5_num + 1, FLAGS.num_h5_file, lr_, train_loss))
print(
'train_r_loss = %.4f, train_g_loss = %.4f , train_b_loss = %.4f, train_rk_loss = %.4f, train_gk_loss = %.4f , train_bk_loss = %.4f' %
(train_r_loss.mean(), train_g_loss.mean(), train_b_loss.mean(), train_rk_loss.mean(),
train_gk_loss.mean(), train_bk_loss.mean()))
model_name = 'model-epoch' # save model
save_path_full = os.path.join(save_path, model_name)
saver.save(sess, save_path_full, global_step=h5_num)
Training_Loss /= (data_size * FLAGS.num_h5_file)
Validation_Loss = sess.run(loss, feed_dict={images: validation_data[0:FLAGS.batch_size, :, :, :],
labels: validation_label[0:FLAGS.batch_size, :, :,
:], is_training: False}) # validation loss
print('%d epoch is finished, Training_Loss = %.4f, Validation_Loss = %.4f' % (
j + 1, Training_Loss, Validation_Loss))