-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathtrain_net.py
executable file
·164 lines (122 loc) · 6.1 KB
/
train_net.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
#!/usr/bin/env python
from SAG_network import *
from timer import Timer
from utils import *
import os
import time
import numpy as np
c_dict = load_from_yml("config.yml")
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(map(str, c_dict['TRAIN']['GPU_ID']))
def snapshot(sess, saver, output_dir, cur_iter, cur_shape, cur_time):
"""Take a snapshot of the network and save the trainable parameters."""
if not os.path.exists(output_dir):
os.makedirs(output_dir)
cur_dir_name = "%s_%s" % (cur_shape, cur_time)
if cur_dir_name is not None:
current_output_dir = output_dir + cur_dir_name + "/"
if not os.path.exists(current_output_dir):
os.makedirs(current_output_dir)
filename = os.path.join(current_output_dir, '%s_%i.ckpt' % (cur_shape, cur_iter))
else:
filename = os.path.join(output_dir, '%s_%i.ckpt' % (cur_shape, cur_iter))
saver.save(sess, filename)
print 'Wrote snapshot to: {:s}'.format(filename)
def train_net(sess, config_dict):
config_dict['TRAIN']['NUM_GPUS'] = len(config_dict['TRAIN']['GPU_ID'])
shape_name = config_dict['TRAIN']['SHAPE_NAME']
max_part_size = 0
if shape_name == 'motorbike':
max_part_size = 5 # motorbike
elif shape_name == 'chair':
max_part_size = 5 # chair
elif shape_name == 'airplane':
max_part_size = 6 # airplane
elif shape_name == 'guitar':
max_part_size = 3 # guitar
elif shape_name == 'lamp':
max_part_size = 4 # lamp
elif shape_name == 'toy_examples':
max_part_size = 2 # toy_examples
config_dict['MAX_PART_SIZE'] = max_part_size
# Construct and initialize a data_runner
data_helper = data_runner(config_dict, for_training=True)
inputs = data_helper.get_placeholders()
# set the random seed
random_seed = config_dict['TRAIN']['RANDOM_SEED']
tf.set_random_seed(random_seed)
np.random.seed(random_seed)
inputs['config_dict'] = config_dict
total_timer = Timer()
total_timer.tic()
cur_time = time.strftime("%Y_%m_%d_%H_%M_%S", time.localtime())
cur_dir_name = "%s_%s" % (shape_name, cur_time)
train_net = SAGNet(inputs, data_helper=data_helper, for_training=True)
print "setup GCDNet....................................................."
train_net.setup()
total_time_diff = total_timer.toc()
print "------------------------------------------------------------------------------"
for variable in tf.trainable_variables():
shape = variable.get_shape()
print "shape:" + str(shape)
print "name:" + variable.name
print "Initial time: step 1: build the network: %g" % (total_time_diff)
results_dir = config_dict['TRAIN']['RESULTS_DIRECTORY']
results_dir = os.path.join(results_dir, cur_dir_name)
model_dir = config_dict['TRAIN']['MODEL_DIRECTORY']
log_dir = config_dict['TRAIN']['LOG_DIRECTORY']
checkpoint_path = config_dict['TRAIN']['PRETRAINED_MODEL_PATH']
snapshot_freq = config_dict['TRAIN']['SNAPSHOT_FREQ']
summary_freq = config_dict['TRAIN']['SUMMARY_FREQ']
current_log_dir = os.path.join(log_dir, cur_dir_name)
train_writer = tf.summary.FileWriter(current_log_dir, sess.graph)
# write the model info to file
model_info_dir = os.path.join(model_dir, cur_dir_name)
data_helper.write_model_info_to_file(model_info_dir)
iter_timer = Timer()
saver = tf.train.Saver(tf.trainable_variables())
sess.run(tf.global_variables_initializer())
if checkpoint_path != '' and checkpoint_path.endswith('.ckpt'):
print "Restore model from checkpoints..."
saver.restore(sess, checkpoint_path)
print "Restore done."
restore_epoch = os.path.splitext(os.path.basename(checkpoint_path))[0]
restore_epoch = int(restore_epoch.split('_')[-1])
print "Restore done. Current starting iteration: %d" % restore_epoch
else:
restore_epoch = 0
# model_vars = tf.trainable_variables()
# tf.contrib.slim.model_analyzer.analyze_vars(model_vars, print_info=True)
current_iter = -1
last_snapshot_iter = -1
n_epochs = config_dict['TRAIN']['ITER_NUM']
for cur_iter in range(restore_epoch, n_epochs):
current_iter = cur_iter
iter_timer.tic()
f_dict = data_helper.get_next_minibatch(cur_iter=cur_iter) # Load the input data for next mini-batch
if cur_iter == 0:
print "train neural network---------------------------------------------------"
if (cur_iter + 1) % summary_freq == 0 or cur_iter < 20:
_, summaries = train_net.train(sess, f_dict, cur_iter, is_summary=True)
train_writer.add_summary(summaries, cur_iter)
else:
_ = train_net.train(sess, f_dict, cur_iter, is_summary=False)
iter_diff = iter_timer.toc()
print "Current iter: %d, time diff: %g" % (cur_iter, iter_diff)
if (cur_iter + 1) % snapshot_freq == 0:
last_snapshot_iter = cur_iter
snapshot(sess, saver, model_dir, cur_iter, cur_shape=shape_name, cur_time=cur_time)
voxels_list, bboxs_list, part_visible_masks = train_net.get_batch_info(sess, f_dict)
data_helper.write_output_to_file(voxels_list=voxels_list, bboxs_list=bboxs_list,
part_visible_masks=part_visible_masks,
input_info_dict=f_dict, output_dir=results_dir,
iter_n=current_iter)
if last_snapshot_iter != current_iter:
snapshot(sess, saver, model_dir, current_iter, cur_shape=shape_name, cur_time=cur_time)
f_dict = data_helper.get_next_minibatch(cur_iter=current_iter)
voxels_list, bboxs_list, part_visible_masks = train_net.get_batch_info(sess, f_dict)
data_helper.write_output_to_file(voxels_list=voxels_list, bboxs_list=bboxs_list,
part_visible_masks=part_visible_masks,
input_info_dict=f_dict, output_dir=results_dir, iter_n=current_iter)
if __name__ == "__main__":
with tf.Session() as sess:
train_net(sess=sess, config_dict=c_dict)