-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtrain.py
337 lines (265 loc) · 13.9 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
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
"""
Copyright (c) College of Mechatronics and Control Engineering, Shenzhen University.
All rights reserved.
Description :
Author：Team Li
"""
from nets.catch_net import factory
from utils import net_tools
from utils import data_pileline_tools
from utils.common_tools import *
from utils.tf_extended import tf_utils
import config
from dataset import dataset_factory
from time import time
import os
import numpy as np
import tensorflow as tf
slim = tf.contrib.slim
tf.app.flags.DEFINE_string(
'backbone_name', 'mobilenet_v2',
'The name of the architecture to train.')
tf.app.flags.DEFINE_float('learning_rate', 1e-3, 'Initial learning rate.')
tf.app.flags.DEFINE_integer(
'batch_size', 20, 'The number of samples in each batch.')
tf.app.flags.DEFINE_integer(
'num_readers', 4,
'The number of parallel readers that read data from the dataset.')
tf.app.flags.DEFINE_integer(
'num_preprocessing_threads', 4,
'The number of threads used to create the batches.')
tf.app.flags.DEFINE_string(
'checkpoint_all', None,
'checkpoint(for all net) full name from which to fine-tune.')
tf.app.flags.DEFINE_string(
'checkpoint_refine', 'checkpoint/mbn_none53x35/refine/mobilenet_v2.model',
'checkpoint(for all net) full name from which to fine-tune.')
tf.app.flags.DEFINE_string(
'train_dir', 'checkpoint/',
'Directory where checkpoints and event logs are written to.')
tf.app.flags.DEFINE_string(
'summary_dir', 'summary/',
'Directory where summary are written to.')
tf.app.flags.DEFINE_integer('max_number_of_steps', None,
'The maximum number of training steps.')
tf.app.flags.DEFINE_integer(
'log_every_n_steps', 20,
'The frequency with which logs are print.')
tf.app.flags.DEFINE_integer(
'summary_every_n_steps', 20,
'The frequency with which summary are writed.')
tf.app.flags.DEFINE_integer(
'save_every_n_steps', 2000,
'The frequency with which model are saved.')
tf.app.flags.DEFINE_boolean(
'fix_refine', True,
'whether fix refine net')
FLAGS = tf.app.flags.FLAGS
DTYPE = tf.float32
global_step = tf.Variable(0, trainable=False, name='global_step')
def main(_):
## assert ##
logger.info('Asserting parameters')
assert FLAGS.batch_size > 0
assert FLAGS.learning_rate >= 0.
assert (FLAGS.log_every_n_steps > 0 or FLAGS.log_every_n_steps == None)
assert (FLAGS.summary_every_n_steps > 0 or FLAGS.summary_every_n_steps == None)
assert (FLAGS.save_every_n_steps > 0 or FLAGS.save_every_n_steps == None)
assert FLAGS.backbone_name in config.supported_backbone_name
## translate the anchor box config to x,y,h,w in all layers ##
layer_n = len(list(config.extract_feat_name[FLAGS.backbone_name]))
anchors_all = net_tools.anchors_all_layer(config.img_size,
config.feat_size_all_layers[FLAGS.backbone_name],
net_tools.init_anchor(layer_n))
## building data pileline ##
logger.info('Building data pileline, using dataset---%s' % ('bdd100k_train'))
with tf.device('/cpu:0'): ## use cpu to read data and batch data
dataset = dataset_factory.get_dataset(
'bdd100k', 'train', './dataset/bdd100k_TfRecord/')
img, labels, bboxes = data_pileline_tools.prepare_data_train(dataset, num_readers=FLAGS.num_readers,
batch_size=FLAGS.batch_size, shuffle=True)
center_bboxes = cornerBboxes_2_centerBboxes(bboxes)
method = config.refine_method.JACCARD_BIGGER
refine_gt, refine_cbboxes, refine_labels, refine_pos_mask = \
net_tools.refine_groundtruth(anchors_all, center_bboxes, labels, method)
list_shape = [1] + [layer_n] * 4
batch_info_for_refine = tf.train.batch(
tf_utils.reshape_list([img, refine_gt, refine_cbboxes, refine_labels, refine_pos_mask]),
batch_size=FLAGS.batch_size,
num_threads=FLAGS.num_preprocessing_threads,
capacity=5 * FLAGS.batch_size)
## the batch img, gt for loss1, and responsible index ##
imgs, refine_gt, refine_cbboxes, refine_labels, refine_pos_mask = \
tf_utils.reshape_list(batch_info_for_refine, list_shape)
imgs = (2.0 / 255.0) * imgs - 1.0
imgs = tf.cast(imgs, dtype=DTYPE)
logger.info('Building model, using backbone---%s' % (FLAGS.backbone_name))
config_dict = {'train_range': config.train_range.REFINE,
'process_backbone_method': config.process_backbone_method.NONE,
'deconv_method': config.deconv_method.LEARN_HALF,
'merge_method': config.merge_method.ADD}
net = factory(inputs=imgs, backbone_name=FLAGS.backbone_name,
is_training=True, dtype=DTYPE, config_dict=config_dict)
logger.info('Total trainable parameters:%s' %
str(np.sum([np.prod(v.get_shape().as_list()) for v in tf.trainable_variables()])))
if config_dict['train_range'] is config.train_range.ALL:
refine_out, det_out, clf_out = net.get_output()
## build refine loss ##
refine_loss = net_tools.refine_loss(refine_out, refine_gt, refine_pos_mask, dtype=DTYPE)
## calculate the groudtruth of offset and classification ##
det_gt, det_pos_mask, det_labels, iou_all_layers = \
net_tools.det_groundtruth(refine_out, refine_gt, refine_cbboxes, refine_labels,
refine_pos_mask, anchors_all)
det_loss, clf_loss = net_tools.det_clf_loss(refine_out, clf_out, det_out, det_gt,
det_pos_mask, det_labels, iou_all_layers, dtype=DTYPE)
## reuse refine net ##
reuse_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
scope="backbone.+|refine.+")
reuse_vars_dict = dict([(var.op.name, var) for var in reuse_vars])
restore_saver = tf.train.Saver(reuse_vars_dict)
if FLAGS.fix_refine:
train_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
scope="^((?!(backbone|refine)).)*$") ##filter the refine model's vars
total_loss = det_loss + clf_loss
else:
train_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
total_loss = refine_loss + det_loss + clf_loss
train_ops = net_tools.optimizer(total_loss, global_step, learning_rate=FLAGS.learning_rate,
batch_szie=FLAGS.batch_size, var_list=train_vars,
fix_learning_rate=False)
## merged the summary op and save the graph##
summary_ops = tf.summary.merge_all()
## saver
saver = tf.train.Saver(tf.global_variables())
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
## create a summary writer ##
writer = tf.summary.FileWriter(FLAGS.summary_dir, sess.graph)
if FLAGS.checkpoint_all == None:
sess.run(tf.global_variables_initializer())
logger.info('TF variables init success...')
else:
tf.train.Saver().restore(sess, FLAGS.checkpoint_all)
logger.info('Load checkpoint for all net success...')
if FLAGS.checkpoint_refine != None:
restore_saver.restore(sess, FLAGS.checkpoint_refine)
logger.info('Load checkpoint for refine net success...')
# start queue
coord = tf.train.Coordinator()
# start the queues #
threads = tf.train.start_queue_runners(coord=coord)
avg_loss = 0.
avg_clf_loss = 0.
avg_det_loss = 0.
avg_time = 0.
tf.get_default_graph().finalize()
while (True):
start = time()
update, summary, t_loss, c_loss, d_loss, current_step= \
sess.run([train_ops, summary_ops, total_loss, clf_loss, det_loss, global_step])
t = round(time() - start, 3)
## for logging
if FLAGS.log_every_n_steps != None:
## caculate average loss ##
step = current_step % FLAGS.log_every_n_steps
avg_loss = (avg_loss * step + t_loss) / (step + 1.)
avg_clf_loss = (avg_clf_loss * step + c_loss) / (step + 1.)
avg_det_loss = (avg_det_loss * step + d_loss) / (step + 1.)
avg_time = (avg_time * step + t) / (step + 1.)
if current_step % FLAGS.log_every_n_steps == FLAGS.log_every_n_steps - 1:
## print info ##
logger.info('Step%s total_loss:%s det_loss:%s clf_loss:%s time_each_step:%s' % \
(str(current_step + 1), str(avg_loss), str(avg_det_loss), str(avg_clf_loss),
str(avg_time)))
avg_loss = 0.
avg_clf_loss = 0.
avg_det_loss = 0.
avg_time = 0.
## for summary
if FLAGS.summary_every_n_steps != None:
if current_step % FLAGS.summary_every_n_steps == FLAGS.summary_every_n_steps - 1:
writer.add_summary(summary, current_step)
if FLAGS.save_every_n_steps != None:
if current_step % FLAGS.save_every_n_steps == FLAGS.save_every_n_steps - 1:
## save model ##
logger.info('Saving model...')
model_name = os.path.join(FLAGS.train_dir, FLAGS.backbone_name + '.model')
saver.save(sess, model_name)
logger.info('Save model sucess...')
if FLAGS.max_number_of_steps != None:
if current_step >= FLAGS.max_number_of_steps:
logger.info('Exit training...')
break
elif config_dict['train_range'] is config.train_range.REFINE:
## get refine output ##
refine_out = net.get_output()
## build refine loss ##
refine_loss = net_tools.refine_loss(refine_out, refine_gt, refine_pos_mask, dtype=DTYPE)
## build optimizer ##
train_ops = net_tools.optimizer(refine_loss, global_step, learning_rate=FLAGS.learning_rate,
batch_szie=FLAGS.batch_size, fix_learning_rate=False)
## merged the summary op and save the graph##
summary_ops = tf.summary.merge_all()
# slim.learning.train(train_op=train_ops, logdir=FLAGS.train_dir, summary_op=merged,
# number_of_steps=FLAGS.max_number_of_steps, log_every_n_steps=FLAGS.log_every_n_steps,
# save_summaries_secs=FLAGS.save_summaries_secs, save_interval_secs=FLAGS.save_interval_secs)
## saver
saver = tf.train.Saver(tf.global_variables())
init = tf.global_variables_initializer()
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
## create a summary writer ##
writer = tf.summary.FileWriter(FLAGS.summary_dir, sess.graph)
if FLAGS.checkpoint_refine == None:
sess.run(init)
logger.info('TF variables init success...')
else:
tf.train.Saver().restore(sess, FLAGS.checkpoint_refine)
logger.info('Load checkpoint success...')
# start queue
coord = tf.train.Coordinator()
# start the queues #
threads = tf.train.start_queue_runners(coord=coord)
avg_refine_loss = 0.
avg_time = 0.
while (True):
start = time()
update, summary, r_loss, current_step = sess.run([train_ops, summary_ops, refine_loss, global_step])
t = round(time() - start, 3)
## for logging
if FLAGS.log_every_n_steps != None:
## caculate average loss ##
step = current_step % FLAGS.log_every_n_steps
avg_refine_loss = (avg_refine_loss * step + r_loss) / (step + 1.)
avg_time = (avg_time * step + t) / (step + 1.)
if current_step % FLAGS.log_every_n_steps == FLAGS.log_every_n_steps - 1:
## print info ##
logger.info('Step_%s refine_loss:%s time:%s' % (str(current_step+1),
str(avg_refine_loss),
str(avg_time)))
avg_refine_loss = 0.
## for summary
if FLAGS.summary_every_n_steps != None:
if current_step % FLAGS.summary_every_n_steps == FLAGS.summary_every_n_steps - 1:
writer.add_summary(summary, current_step)
if FLAGS.save_every_n_steps != None:
if current_step % FLAGS.save_every_n_steps == FLAGS.save_every_n_steps - 1:
## save model ##
logger.info('Saving model...')
model_name = os.path.join(FLAGS.train_dir, FLAGS.backbone_name + '.model')
saver.save(sess, model_name)
logger.info('Save model sucess...')
if FLAGS.max_number_of_steps != None:
if current_step >= FLAGS.max_number_of_steps:
logger.info('Exit training...')
break
# terminate the threads #
coord.request_stop()
coord.join(threads)
else:
raise ValueError("error")
pass
if __name__ == '__main__':
tf.app.run()