-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
207 lines (153 loc) · 7.01 KB
/
train.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
"""
Copyright (c) College of Mechatronics and Control Engineering, Shenzhen University.
All rights reserved.
Description :
Author：Team Li
"""
import tensorflow as tf
import numpy as np
import logging
import os
from net.u_net import u_net
from net.fpn_net import fpn_net
from net.fpn_net_lite import fpn_net_lite
from dataset.bdd_daytime import bdd_daytime
tf.app.flags.DEFINE_string(
'checkpoint_dir', '',
'The path to a checkpoint from which to fine-tune.')
tf.app.flags.DEFINE_string(
'train_dir', './checkpoint/fpn_lite',
'Directory where checkpoints are written to.')
tf.app.flags.DEFINE_string(
'summary_dir', './summary/fpn_lite',
'Directory where checkpoints are written to.')
tf.app.flags.DEFINE_float('learning_rate', 1e-3, 'Initial learning rate.')
tf.app.flags.DEFINE_integer(
'batch_size', 6, 'The number of samples in each batch.')
tf.app.flags.DEFINE_integer(
'f_log_step', 20,
'The frequency with which logs are print.')
tf.app.flags.DEFINE_integer(
'f_summary_step', 20,
'The frequency with which the model is saved, in step.')
tf.app.flags.DEFINE_integer(
'f_save_step', 9999,
'The frequency with which summaries are saved, in step.')
tf.app.flags.DEFINE_integer(
'f_eval_step', 20,
'The frequency with which summaries are saved, in step.')
tf.app.flags.DEFINE_integer(
'max_step', 50000,
'The frequency with which summaries are saved, in step.')
FLAGS = tf.app.flags.FLAGS
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
input = tf.placeholder(shape=[None, 278, 418, 3], dtype=tf.float32)
groundtruth = tf.placeholder(shape=[None, 278, 418, 3], dtype=tf.float32)
global_step = tf.Variable(0, trainable=False, name='global_step')
lr = tf.placeholder(dtype=tf.float32)
def _smooth_l1(x):
"""Smoothed absolute function. Useful to compute an L1 smooth error.
Define as:
x^2 / 2 if abs(x) < 1
abs(x) - 0.5 if abs(x) > 1
We use here a differentiable definition using min(x) and abs(x). Clearly
not optimal, but good enough for our purpose!
"""
absx = tf.abs(x)
minx = tf.minimum(absx, 1)
r = 0.5 * ((absx - 1) * minx + absx) ## smooth_l1
return r
def build_graph(input):
output, attention_pairs, attentions = fpn_net_lite(input, is_training=True)
attention_decay = 5e-2
attention_regularization = 0.
for attention_pair in attention_pairs:
attention_regularization += tf.reduce_mean(_smooth_l1(attention_pair[0] - attention_pair[1]))
tf.summary.scalar('attention_regularization', attention_regularization)
## loss_1
mse_loss = tf.reduce_sum(_smooth_l1(output - groundtruth)) / FLAGS.batch_size
output = (output + 1.) * 255. / 2.
gt_img = (groundtruth + 1.) * 255. / 2.
psnr = tf.image.psnr(output, gt_img, max_val=255.)
psnr_loss = tf.reduce_sum(1 / (psnr + 1e-8))
ssmi_loss = tf.image.ssim_multiscale(output, gt_img, max_val=255.)
ssmi_loss = tf.maximum(ssmi_loss, 1e-8)
ssmi_loss = tf.reduce_sum(-tf.log(ssmi_loss))
# LOSS_MSE #
loss_1 = mse_loss
# LOSS_SSMI_MSE #
loss_2 = ssmi_loss*10000 + mse_loss
# LOSS_SSMI_PSNR RECOMMEND#
loss_3 = ssmi_loss + psnr_loss + attention_decay*attention_regularization##
tf.summary.scalar('total_loss', loss_3)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
optimizer = tf.train.AdamOptimizer(learning_rate=lr)
grads_and_vars = optimizer.compute_gradients(loss_3)
## clip the gradients ##
capped_gvs = [(tf.clip_by_value(grad, -5., 5.), var)
for grad, var in grads_and_vars]
train_op = optimizer.apply_gradients(capped_gvs, global_step=global_step)
return output, loss_3, train_op
def main(_):
output, loss, train_op = build_graph(input)
logger.info('Total trainable parameters:%s'%(str(np.sum([np.prod(v.get_shape().as_list()) \
for v in tf.trainable_variables()]))))
saver = tf.train.Saver(max_to_keep=5)
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
init = tf.global_variables_initializer()
merge_ops = tf.summary.merge_all()
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
## create a summary writer ##
summary_dir = os.path.join(FLAGS.summary_dir)
writer = tf.summary.FileWriter(FLAGS.summary_dir, sess.graph)
pd = bdd_daytime(batch_size=FLAGS.batch_size, for_what='train', shuffle=True)
if ckpt:
logger.info('loading %s...'%str(ckpt.model_checkpoint_path))
saver.restore(sess, ckpt.model_checkpoint_path)
logger.info('load %s success...'%str(ckpt.model_checkpoint_path))
else:
sess.run(init)
logger.info('Init Tf parameters success...')
avg_loss = 0.
current_step = sess.run(global_step)
while current_step < FLAGS.max_step:
if current_step<FLAGS.max_step//3:
learning_rate = FLAGS.learning_rate
elif current_step<FLAGS.max_step*2//3:
learning_rate = FLAGS.learning_rate / 10.
else:
learning_rate = FLAGS.learning_rate / 10.
gt_imgs, train_imgs = pd.load_batch()
update_op, m_ops, l, current_step = sess.run([train_op, merge_ops, loss, global_step],
feed_dict={input:train_imgs,
groundtruth:gt_imgs,
lr:learning_rate})
if FLAGS.f_log_step != None:
## caculate average loss ##
step = current_step % FLAGS.f_log_step
avg_loss = (avg_loss * step + l) / (step + 1.)
if current_step % FLAGS.f_log_step == FLAGS.f_log_step - 1:
## print info ##
logger.info('Step%s loss:%s' % (str(current_step), str(avg_loss)))
avg_loss = 0.
if FLAGS.f_summary_step != None:
if current_step % FLAGS.f_summary_step == FLAGS.f_summary_step - 1:
## summary ##
writer.add_summary(m_ops, current_step)
if FLAGS.f_save_step != None:
if current_step % FLAGS.f_save_step == FLAGS.f_save_step - 1:
## save model ##
logger.info('Saving model...')
model_name = os.path.join(FLAGS.train_dir, 'dark_aug.model')
saver.save(sess, model_name, global_step=current_step)
logger.info('Save model sucess...')
logger.info('Saving model...')
model_name = os.path.join(FLAGS.train_dir, 'dark_aug_final.model')
saver.save(sess, model_name, global_step=current_step)
logger.info('Save model sucess...')
if __name__ == '__main__':
tf.app.run()