-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathtrain.py
150 lines (134 loc) · 6.29 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
# -*- coding: utf-8 -*-
from __future__ import division
import os
import argparse
import tensorflow as tf
from utils.model_utils import make_dir
from dataset.dataset import ImageDataSet
from model.model import yolo_net, yolo_loss
parser = argparse.ArgumentParser()
parser.add_argument("--load_weights", type=str, default='False', help="Whether to load weights, True or False")
parser.add_argument("--batch_size", type=int, default=8, help="Set the batch_size")
parser.add_argument("--weights_path", type=str, default="./weights/...", help="Set the weights_path")
parser.add_argument("--save_dir", type=str, default="./weights/", help="Dir to save weights")
parser.add_argument("--gpu_id", type=str, default='0', help="Specify GPU device")
parser.add_argument("--num_iter", type=int, default=16000, help="num_max_iter")
parser.add_argument("--save_interval", type=int, default=2, help="Save once every two epochs")
args = parser.parse_args()
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
SCALE = 32
GRID_W, GRID_H = 32, 24
N_CLASSES = 8
N_ANCHORS = 5
IMAGE_HEIGHT, IMAGE_WIDTH, IMAGE_DEPTH = GRID_H * SCALE, GRID_W * SCALE, 3
batch_size = args.batch_size
train_dataset = ImageDataSet(data_set='train',
mode='train',
load_to_memory=False)
test_dataset = ImageDataSet(data_set='test',
mode='test',
flip=False,
aug_hsv=False,
random_scale=False,
load_to_memory=False)
num_val_step = int(test_dataset.num_samples / args.batch_size)
save_steps = int(train_dataset.num_samples / args.batch_size * args.save_interval)
def print_info():
print("train samples: {}".format(train_dataset.num_samples))
print("test samples: {}".format(test_dataset.num_samples))
print("batch_size: {}".format(args.batch_size))
print("iter steps: {}".format(args.num_iter))
def train(load_weights='False'):
make_dir(args.save_dir)
max_val_loss = 99999999.0
global_step = tf.Variable(0, trainable=False)
learning_rate = tf.train.exponential_decay(0.001,
global_step,
1500,
0.96,
staircase=True)
image = tf.placeholder(
shape=[None, IMAGE_HEIGHT, IMAGE_WIDTH, IMAGE_DEPTH],
dtype=tf.float32,
name='image_placeholder')
label = tf.placeholder(shape=[None, GRID_H, GRID_W, N_ANCHORS, 8],
dtype=tf.float32,
name='label_placeholder')
train_flag = tf.placeholder(dtype=tf.bool, name='flag_placeholder')
with tf.variable_scope('net'):
y = yolo_net(image, train_flag)
with tf.name_scope('loss'):
loss, loss_xy, loss_wh, loss_re, loss_im, loss_obj, loss_no_obj, loss_c = yolo_loss(
y, label, batch_size)
loss_xy_sum = tf.summary.scalar("loss_xy_sum", loss_xy)
loss_wh_sum = tf.summary.scalar("loss_wh_sum", loss_wh)
loss_re_sum = tf.summary.scalar("loss_re_sum", loss_re)
loss_im_sum = tf.summary.scalar("loss_im_sum", loss_im)
loss_obj_sum = tf.summary.scalar("loss_obj_sum", loss_obj)
loss_no_obj_sum = tf.summary.scalar("loss_no_obj_sum", loss_no_obj)
loss_c_sum = tf.summary.scalar("loss_c", loss_c)
loss_sum = tf.summary.scalar("loss", loss)
loss_tensorboard_sum = tf.summary.merge([
loss_xy_sum, loss_wh_sum, loss_re_sum, loss_im_sum,
loss_obj_sum, loss_no_obj_sum, loss_c_sum, loss_sum
])
opt = tf.train.AdamOptimizer(learning_rate=learning_rate)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
train_step = opt.minimize(loss, global_step=global_step)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
writer = tf.summary.FileWriter("./logs", sess.graph)
if load_weights == 'True':
print("load weights from {}".format(args.weights_path))
saver = tf.train.import_meta_graph(args.weights_path + '.meta')
saver.restore(sess, args.weights_path)
print('load weights done!')
for step, (train_image_data, train_label_data) in enumerate(
train_dataset.get_batch(batch_size)):
_, lr, train_loss, data, summary_str = sess.run(
[train_step, learning_rate, loss, y, loss_tensorboard_sum],
feed_dict={
train_flag: True,
image: train_image_data,
label: train_label_data
})
writer.add_summary(summary_str, step)
if step % 10 == 0:
print('iter: %i, loss: %f, lr: %f' % (step, train_loss, lr))
if (step + 1) % save_steps == 0:
print("val...")
val_loss = 0.0
for val_step, (val_image_data, val_label_data) in enumerate(
test_dataset.get_batch(batch_size)):
val_loss += sess.run(loss,
feed_dict={
train_flag: False,
image: val_image_data,
label: val_label_data
})
if val_step + 1 == num_val_step:
break
val_loss /= num_val_step
print("iter: {} val_loss: {:.2f}".format(step, val_loss))
if val_loss < max_val_loss:
saver.save(sess,
os.path.join(
args.save_dir,
'Complex_YOLO_train_loss_{:.2f}_val_loss_{:.2f}'.format(
train_loss, val_loss)),
global_step=global_step)
max_val_loss = val_loss
if step + 1 == args.num_iter:
saver.save(sess,
os.path.join(
args.save_dir,
'Complex_YOLO_final_train_loss_{:.2f}'.format(
train_loss)),
global_step=global_step)
print("training done!")
break
if __name__ == "__main__":
print_info()
train(load_weights=args.load_weights)