-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainVGG.py
141 lines (130 loc) · 7.99 KB
/
trainVGG.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
from vgg16 import vgg16
import cifar10_input as loader
import tensorflow as tf
import numpy as np
import resource, sys
from datetime import datetime
import argparse
def main():
# _IMG_SIZE = 32
_IMG_SIZE = 28
_IMG_CHANNEL = 3
_IMG_CLASS = 10
parser = argparse.ArgumentParser(description='vgg16')
parser.add_argument('--data_dir', type=str, default='./cifar-10-batches-py/')
parser.add_argument('--learning_rate', type=float, default=0.0001)
parser.add_argument('--batch_size', type=int, default=100) # must be integer times of total images_num
parser.add_argument('--summary_dir', type=str, default='./summary/vgglog/')
parser.add_argument('--max_epoch', type=int, default=60)
parser.add_argument('--eval_freq', type=int, default=1)
parser.add_argument('--save_freq', type=int, default=300)
args = parser.parse_args()
print(args)
with tf.device('/gpu:1'):
# -----------------------------------------------------------------------------
# BUILD GRAPH
# -----------------------------------------------------------------------------
inputs = tf.placeholder(dtype=tf.float32, shape=[None, _IMG_SIZE, _IMG_SIZE, _IMG_CHANNEL], name='inputs')
labels = tf.placeholder(dtype=tf.int64, shape=None, name='labels')
learning_rate = tf.placeholder(dtype=tf.float32, shape=[], name='learning_rate')
is_training = tf.placeholder(dtype=tf.bool, shape=[], name='is_training')
keepPro = tf.placeholder(dtype=tf.float32, shape=[], name='keep_prob')
model = vgg16(imgs=inputs, weights=args.data_dir + 'vgg16_weights.npz', is_training=is_training, keepPro=keepPro)
logits = model.probs
# loss
with tf.name_scope('loss'):
loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits,
labels=labels,
name='cross_entropy'))
tf.summary.scalar('loss', loss)
# optimize
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(extra_update_ops):
with tf.name_scope("train"):
optimizer = tf.train.RMSPropOptimizer(learning_rate=learning_rate,
decay=0.9,
momentum=0.0,
epsilon=1e-10,
use_locking=False,
name='RMSProp').minimize(loss)
# evaluate acc
with tf.name_scope('accuracy'):
accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(logits, axis=1), labels), tf.float32))
tf.summary.scalar('accuracy', accuracy)
# ready for summary or save
merged = tf.summary.merge_all()
saver = tf.train.Saver()
print("[BUILD GRAPH] memory_usage=%f" % (resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024), file=sys.stderr)
# -----------------------------------------------------------------------------
# LOAD DATA
# -----------------------------------------------------------------------------
train_images, train_labels, test_images, test_labels = loader.load_batch_data_aug(args.data_dir, args.batch_size)
print("[LOAD DATA] memory_usage=%f" % (resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024), file=sys.stderr)
# -----------------------------------------------------------------------------
# START THE SESSION
# -----------------------------------------------------------------------------
cur_lr = args.learning_rate # current learning rate
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
train_writer = tf.summary.FileWriter(logdir=args.summary_dir + datetime.now().strftime('%Y%m%d-%H%M%S') + '/train/',
graph=sess.graph)
test_writer = tf.summary.FileWriter(logdir=args.summary_dir + datetime.now().strftime('%Y%m%d-%H%M%S') + '/test/')
for epoch in range(args.max_epoch):
train_accs = []
train_losses = []
# train
for index, images_batch in enumerate(train_images):
_, summary, train_loss, train_acc = sess.run(fetches = [optimizer, merged, loss, accuracy],
feed_dict = { inputs: images_batch,
labels: train_labels[index],
learning_rate: cur_lr,
is_training: True,
keepPro: 0.6})
train_accs.append(train_acc)
train_losses.append(train_loss)
# print('[batch] {} done'.format(index), end='\r')
train_avg_acc = float(np.mean(np.asarray(train_accs)))
train_avg_loss = float(np.mean(np.asarray(train_losses)))
train_summary = tf.Summary(value=[tf.Summary.Value(tag="accuracy", simple_value=train_avg_acc),
tf.Summary.Value(tag="loss", simple_value=train_avg_loss),
tf.Summary.Value(tag="learning_rate", simple_value=cur_lr)])
train_writer.add_summary(summary, epoch)
train_writer.add_summary(train_summary, epoch)
print('=' * 20 + 'EPOCH {} [TRAIN]'.format(epoch) + '=' * 20)
print('acc: {0}, loss: {1}'.format(train_avg_acc, train_avg_loss))
# evaluate
if (epoch + 1) % args.eval_freq == 0:
test_accs = []
test_losses = []
for index, test_images_batch in enumerate(test_images):
test_loss, test_acc = sess.run(fetches = [loss, accuracy],
feed_dict = { inputs: test_images_batch,
labels: test_labels[index],
learning_rate: cur_lr,
is_training: False,
keepPro: 0.6})
test_accs.append(test_acc)
test_losses.append(test_loss)
test_avg_acc = float(np.mean(np.asarray(test_accs)))
test_avg_loss = float(np.mean(np.asarray(test_losses)))
test_summary = tf.Summary(value=[tf.Summary.Value(tag="accuracy", simple_value=test_avg_acc),
tf.Summary.Value(tag="loss", simple_value=test_avg_loss)])
test_writer.add_summary(test_summary, epoch)
print('=' * 20 + 'EPOCH {} [EVAL]'.format(epoch) + '=' * 20)
print('acc: {0}, loss: {1}'.format(test_avg_acc, test_avg_loss))
# lr decay
cur_lr = lr(cur_lr, epoch)
# save
if (epoch + 1) % args.save_freq == 0:
checkpoint_file = args.summary_dir + 'model.ckpt'
saver.save(sess, checkpoint_file, global_step=epoch)
print('Saved checkpoint')
train_writer.close()
test_writer.close()
def lr(lr, epoch):
lr = lr * 0.9
return lr
if __name__ == '__main__':
main()