-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_inception_v3_hvc_decaying.py
209 lines (177 loc) · 9.36 KB
/
train_inception_v3_hvc_decaying.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
# Copyright 2019 Adam Byerly. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
import argparse
import tensorflow as tf
from datetime import datetime
from input_sieve import DataSet, train_inputs, eval_inputs
from input_sieve import non_blacklisted_eval_inputs
from inception_v3.output import Output
from inception_v3.model_hvc import run_towers
def train(out, sess, epoch, training_steps, train_op, loss_op,
global_step, learning_rate, is_training_ph, is_validating_nbl_ph):
g_step = 0
for i in range(training_steps):
out.train_step_begin(i)
_, l, g_step, lr = sess.run(
[train_op, loss_op, global_step, learning_rate],
feed_dict={is_training_ph: True,
is_validating_nbl_ph: False},
options=out.get_run_options(),
run_metadata=out.get_run_metadata())
out.train_step_end(
sess, epoch, g_step, i, l, lr, training_steps,
feed_dict={is_training_ph: True,
is_validating_nbl_ph: False})
out.train_end(sess, epoch, g_step)
def validate(out, sess, epoch, validation_steps, loss_op,
acc_top_1_op, acc_top_5_op, global_step, learning_rate,
is_training_ph, is_validating_nbl_ph):
g_step, acc_top1, acc_top5, test_loss, lr = (0, 0, 0, 0, 0)
for i in range(validation_steps):
out.validation_step_begin(i, validation_steps)
g_step, l, acc1, acc5, lr = sess.run(
[global_step, loss_op, acc_top_1_op, acc_top_5_op, learning_rate],
feed_dict={is_training_ph: False,
is_validating_nbl_ph: False})
acc_top1 = (acc1 + (i * acc_top1)) / (i + 1)
acc_top5 = (acc5 + (i * acc_top5)) / (i + 1)
test_loss = (l + (i * test_loss)) / (i + 1)
out.validation_end(sess, epoch, g_step, False,
test_loss, lr, acc_top1, acc_top5)
def validate_nbl(out, sess, epoch, nbl_validation_steps, loss_op,
acc_top_1_op, acc_top_5_op, global_step, learning_rate,
is_training_ph, is_validating_nbl_ph):
g_step, acc_top1, acc_top5, test_loss, lr = (0, 0, 0, 0, 0)
for i in range(nbl_validation_steps):
out.validation_step_begin(i, nbl_validation_steps)
g_step, l, acc1, acc5, lr = sess.run(
[global_step, loss_op, acc_top_1_op, acc_top_5_op, learning_rate],
feed_dict={is_training_ph: False,
is_validating_nbl_ph: True})
acc_top1 = (acc1 + (i * acc_top1)) / (i + 1)
acc_top5 = (acc5 + (i * acc_top5)) / (i + 1)
test_loss = (l + (i * test_loss)) / (i + 1)
out.validation_end(sess, epoch, g_step, True,
test_loss, lr, acc_top1, acc_top5)
def go(start_epoch, end_epoch, run_name, weights_file,
profile_compute_time_every_n_steps, save_summary_info_every_n_steps,
log_annotated_images, image_size, batch_size, num_gpus,
data_dir, black_list_file, log_dir, do_validate_all, do_validate_nbl):
tf.reset_default_graph()
out = Output(log_dir, run_name, profile_compute_time_every_n_steps,
save_summary_info_every_n_steps)
############################################################################
# Data feeds
############################################################################
out.log_msg("Setting up data feeds...")
training_dataset = DataSet('train', image_size, batch_size, num_gpus,
data_dir, None)
validation_dataset = DataSet('validation', image_size, batch_size, num_gpus,
data_dir, black_list_file)
training_data = train_inputs(training_dataset, log_annotated_images)
validation_data = eval_inputs(validation_dataset, log_annotated_images)
nbl_val_data = non_blacklisted_eval_inputs(
validation_dataset, log_annotated_images)
training_steps = training_dataset.training_batches_per_epoch()
validation_steps = validation_dataset.validation_batches_per_epoch()
nbl_val_steps = validation_dataset.nbl_validation_batches_per_epoch()
############################################################################
# Tensorflow placeholders and operations
############################################################################
with tf.device("/device:CPU:0"): # set the default device to the CPU
with tf.name_scope("input/placeholders"):
is_training_ph = tf.placeholder(tf.bool)
is_validating_nbl_ph = tf.placeholder(tf.bool)
global_step = tf.train.get_or_create_global_step()
decay_steps = int(training_steps*1.0)
learning_rate_op = tf.train.exponential_decay(0.001,
global_step, decay_steps, 0.96, staircase=True)
learning_rate_op = tf.maximum(learning_rate_op, 1e-6)
opt = tf.train.AdamOptimizer(learning_rate_op)
train_op, loss_op,\
acc_top_1_op, \
acc_top_5_op = run_towers(opt, global_step, is_training_ph,
is_validating_nbl_ph, training_data,
validation_data, nbl_val_data,
DataSet.num_classes(), num_gpus)
out.log_msg("Starting Session...")
############################################################################
# Tensorflow session
############################################################################
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
out.set_session_graph(sess.graph)
if weights_file is not None:
out.log_msg("Restoring weights file: {}".format(weights_file))
tf.train.Saver().restore(sess, weights_file)
else:
tf.global_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
for e in range(start_epoch, end_epoch + 1):
train(out, sess, e, training_steps, train_op, loss_op,
global_step, learning_rate_op,
is_training_ph, is_validating_nbl_ph)
if do_validate_all:
validate(out, sess, e, validation_steps, loss_op,
acc_top_1_op, acc_top_5_op, global_step,
learning_rate_op, is_training_ph,
is_validating_nbl_ph)
if do_validate_nbl:
validate_nbl(out, sess, e, nbl_val_steps, loss_op,
acc_top_1_op, acc_top_5_op, global_step,
learning_rate_op, is_training_ph,
is_validating_nbl_ph)
except tf.errors.OutOfRangeError:
out.log_msg("Finished.")
finally:
coord.request_stop()
coord.join(threads)
sess.close()
out.close_files()
################################################################################
# Entry point
################################################################################
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="inception_v3")
parser.add_argument("-se", "--start_epoch", default=1, type=int)
parser.add_argument("-ee", "--end_epoch", default=175, type=int)
parser.add_argument("-rn", "--run_name",
default=datetime.now().strftime("%Y%m%d%H%M%S"))
parser.add_argument("-wf", "--weights_file", default=None)
parser.add_argument("-pct", "--profile_compute_time_every_n_steps",
default=None, type=int)
parser.add_argument("-ssi", "--save_summary_info_every_n_steps",
default=None, type=int)
parser.add_argument("-lai", "--log_annotated_images",
default=False, type=bool)
parser.add_argument("-is", "--image_size", default=299, type=int)
parser.add_argument("-bs", "--batch_size", default=96, type=int)
parser.add_argument("-g", "--gpus", default=2, type=int)
parser.add_argument("-ld", "--log_dir", default="logs")
parser.add_argument("-dd", "--data_dir",
default="D:\\workingbasedir\\processed")
parser.add_argument("-blf", "--black_list_file",
default="ILSVRC2015_clsloc_validation_blacklist.txt")
parser.add_argument("-vall", "--validate_all", default=True, type=bool)
parser.add_argument("-vnbl", "--validate_nbl", default=True, type=bool)
args = parser.parse_args()
print(args)
go(args.start_epoch, args.end_epoch, args.run_name, args.weights_file,
args.profile_compute_time_every_n_steps,
args.save_summary_info_every_n_steps, args.log_annotated_images,
args.image_size, args.batch_size, args.gpus,
args.data_dir, args.black_list_file, args.log_dir,
args.validate_all, args.validate_nbl)