-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
executable file
·172 lines (155 loc) · 8.55 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
#!/bin/env python
import tensorflow as tf
from tensorflow.python import debug as tf_debug
import os
import random
import numpy as np
from nets.inception_resnet_v2 import inception_resnet_v2, inception_resnet_v2_arg_scope
import argparse
import common
import time
import cv2
import logging
import csv
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
slim = tf.contrib.slim
ap = argparse.ArgumentParser()
ap.add_argument('--resume', action='store_true', default=False)
ap.add_argument('--train_dataset', type=common.readable_directory, required=True)
ap.add_argument('--val_dataset', type=common.readable_directory, required=True)
ap.add_argument('--train_dir', type=common.readable_directory, required=True)
ap.add_argument('--learning_rate', default=1e-4, type=common.positive_float)
ap.add_argument('--epoch', default=5000, type=common.positive_int)
ap.add_argument('--batch_size', default=32, type=common.positive_int)
ap.add_argument('--image_size', default=224, type=common.positive_int)
ap.add_argument('--pretrained_model', default='./pretrained_model/inception_resnet_v2.ckpt', type=common.readable_directory)
args = ap.parse_args()
if not os.path.isdir(args.train_dir):
os.system('mkdir -p {}'.format(args.train_dir))
_RGB_MEAN = [123.68, 116.78, 103.94]
time_identifier = time.strftime("%Y-%m-%d-%H:%M:%S", time.localtime())
log = logging.getLogger('tensorflow')
log.setLevel(logging.INFO)
fh = logging.FileHandler(os.path.join(args.train_dir, '{}_{}.log'.format('log', time_identifier)))
fh.setLevel(logging.INFO)
log.addHandler(fh)
def get_batch(content, is_train):
dataset = tf.data.TextLineDataset(content)
dataset = dataset.map(decode_line).shuffle(buffer_size=1000).batch(args.batch_size)
if is_train:
dataset = dataset.repeat(1)
else:
dataset = dataset.repeat(1)
return dataset.make_initializable_iterator()
def decode_line(line):
dir_, label = tf.decode_csv(records=line, record_defaults=[["string"], ["string"]], field_delim=" ")
label = tf.one_hot(tf.string_to_number(label, tf.int32), depth=2, dtype=tf.float32)
imagecontent = tf.read_file(dir_)
image = tf.image.decode_png(imagecontent, channels=3)
image = tf.cast(image, tf.float32)
image = tf.image.resize_images(image, [args.image_size, args.image_size])
image = (image - _RGB_MEAN) / 128.
return image, label
train_iter = get_batch(content=[args.train_dataset], is_train=True)
val_iter = get_batch(content=[args.val_dataset], is_train=False)
def validation(sess, val_iter):
with tf.contrib.slim.arg_scope(inception_resnet_v2_arg_scope()):
image_op = tf.get_default_graph().get_tensor_by_name('image_input:0')
label_op = tf.get_default_graph().get_tensor_by_name('label_input:0')
acc, op_acc = tf.get_default_graph().get_tensor_by_name('val_acc/value:0'), tf.get_default_graph().get_tensor_by_name('val_acc/update_op:0')
precision, op_precision = tf.get_default_graph().get_tensor_by_name('val_precision/value:0'), tf.get_default_graph().get_tensor_by_name('val_precision/update_op:0')
recall, op_recall = tf.get_default_graph().get_tensor_by_name('val_recall/value:0'), tf.get_default_graph().get_tensor_by_name('val_recall/update_op:0')
image, label = val_iter.get_next()
sess.run(val_iter.initializer)
try:
while True:
_images, _labels = sess.run([image, label])
sess.run([op_acc, op_precision, op_recall], feed_dict={image_op: _images, label_op: _labels})
except tf.errors.OutOfRangeError:
pass
finally:
_acc, _precision, _recall = sess.run([acc, precision, recall])
return _acc, _precision, _recall
with tf.contrib.slim.arg_scope(inception_resnet_v2_arg_scope()):
images = tf.placeholder(tf.float32, (None, args.image_size, args.image_size, 3), name='image_input')
labels = tf.placeholder(tf.float32, (None, 2), name='label_input')
net, endpoints = inception_resnet_v2(images, is_training=True, create_aux_logits=False, num_classes=None)
net = slim.flatten(endpoints['Conv2d_7b_1x1'])
net = slim.dropout(net, 0.8, is_training=True, scope='Dropout2')
logits = tf.contrib.layers.fully_connected(net, 2, activation_fn=None, scope='Logits2')
predictions = tf.nn.softmax(logits, name='Predictions2')
model_variables = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, 'InceptionResnetV2')
global_step = tf.Variable(0, name='global_step', trainable=False)
learning_rate = tf.train.exponential_decay(
learning_rate=args.learning_rate,
global_step=global_step,
decay_steps=1000,
decay_rate=0.96,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
loss = tf.losses.log_loss(labels, predictions, weights=10)
tf.summary.scalar('loss', loss)
metric_acc, op_acc = tf.metrics.accuracy(tf.argmax(labels, 1), tf.argmax(predictions, 1), name='train_acc')
eval_metric_acc, eval_op_acc = tf.metrics.accuracy(tf.argmax(labels, 1), tf.argmax(predictions, 1), name='val_acc')
eval_metric_p, eval_op_p = tf.metrics.precision(tf.argmax(labels, 1), tf.argmax(predictions, 1), name='val_precision')
eval_metric_r, eval_op_r = tf.metrics.recall(tf.argmax(labels, 1), tf.argmax(predictions, 1), name='val_recall')
tf.summary.scalar('train_acc', metric_acc)
tf.summary.scalar('val_acc', eval_metric_acc)
tf.summary.scalar('val_precision', eval_metric_p)
tf.summary.scalar('val_recall', eval_metric_r)
train_op = tf.train.AdamOptimizer(learning_rate).minimize(loss, global_step=global_step)
ckpt_saver = tf.train.Saver(max_to_keep=0)
train_record_fn = os.path.join(args.train_dir, 'train-{}.csv'.format(time_identifier))
val_record_fn = os.path.join(args.train_dir, 'val-{}.csv'.format(time_identifier))
with tf.Session(config=config) as sess, open(train_record_fn, 'w') as ftrain, open(val_record_fn, 'w') as fval:
if args.resume:
saver = tf.train.Saver()
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
last_checkpoint = tf.train.latest_checkpoint(args.train_dir)
saver.restore(sess, last_checkpoint)
tf.logging.info('Resume training from [{}]'.format(last_checkpoint))
else:
saver = tf.train.Saver(model_variables)
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
tf.logging.info('Load pretrained model from [{}]'.format(args.pretrained_model))
saver.restore(sess, args.pretrained_model)
train_writer = csv.writer(ftrain)
train_writer.writerow(['global_step', 'loss', 'accuracy'])
val_writer = csv.writer(fval)
val_writer.writerow(['global_step', 'accuracy', 'precision', 'recall'])
merged_summary = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(args.train_dir, sess.graph)
train_image, train_label = train_iter.get_next()
try:
for _epoch in range(int(args.epoch)):
sess.run(train_iter.initializer)
while True:
try:
start_time = time.time()
_images, _labels = sess.run([train_image, train_label])
_, _lr, _summary, _step, _acc, _loss = sess.run([train_op, learning_rate, merged_summary, global_step, op_acc, loss], feed_dict={images: _images, labels: _labels})
elapsed_time = time.time() - start_time
tf.logging.info("Time: {} | Learning Rate: {} | Epoch: {} | Global_Step: {} | Loss: {} | Accuracy: {}".format(elapsed_time, _lr, _epoch, _step, _loss, _acc))
train_writer.writerow([_step, _loss, _acc])
ftrain.flush()
if _step % 1000 == 0:
ckpt = ckpt_saver.save(sess, os.path.join(args.train_dir, 'model'), global_step=_step)
tf.logging.info('Save ckpt-{} in {}'.format(_step, ckpt))
if _step % 500 == 0:
_eval_acc, _eval_p, _eval_r = validation(sess, val_iter)
val_writer.writerow([_step, _eval_acc, _eval_p, _eval_r])
fval.flush()
tf.logging.info('*'*80)
tf.logging.info("Validation | Global_Step: {} | Accuracy: {} | Precision: {} | Recall: {}".format(_step, _eval_acc, _eval_p, _eval_r))
tf.logging.info('*'*80)
except tf.errors.OutOfRangeError:
break
except tf.errors.OutOfRangeError:
tf.logging.info('Finished')
finally:
ckpt = ckpt_saver.save(sess, os.path.join(args.train_dir, 'model'), global_step=tf.train.get_global_step())
tf.logging.info('Save ckpt in: {}'.format(ckpt))
sess.close()