forked from mariolew/TF-FaceDetection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_net_48.py
executable file
·137 lines (108 loc) · 4.44 KB
/
train_net_48.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
import tensorflow as tf
import numpy as np
from image_inputs import inputs
from models import fcn_48_detect
import h5py
import random
def train():
f = h5py.File('net_48_neg_for_train.hdf5','r')
neg_list = f['imgs']
neg_len = neg_list.len()
lists = ['net_pos_list.txt','net_pos_flip_list.txt']
image_train, label_train = inputs(lists, [48, 48, 3], 20)
net_output = fcn_48_detect(0.3)
global_step = tf.Variable(0, tf.int32)
starter_learning_rate = 0.002
learning_rate = tf.train.exponential_decay(
learning_rate=starter_learning_rate,
global_step=global_step,
decay_steps=1000,
decay_rate=0.99,
staircase=True,
name=None)
opt_vars_48 = [v for v in tf.trainable_variables() if v.name.startswith('net_48')]
train_step = tf.train.AdamOptimizer(learning_rate).minimize(net_output['cost'], var_list=opt_vars_48, global_step=global_step)
sess = tf.Session()
opt_vars_24 = [v for v in tf.trainable_variables() if v.name.startswith('net_24')]
saver_24 = tf.train.Saver(opt_vars_24)
saver_48 = tf.train.Saver()
# import pdb; pdb.set_trace()
sess.run(tf.initialize_all_variables())
coord = tf.train.Coordinator()
# tf.get_default_graph().finalize()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
saver_24.restore(sess, 'model/model_net_24-210000')
try:
for i in range(400000):
# sn_rand = np.random.random_integers(0, neg_len-1, 40)
sn_rand = np.array(random.sample(range(neg_len), 40))
sn_rand = list(np.sort(sn_rand))
imgs_pos = sess.run(image_train)
# import pdb; pdb.set_trace()
try:
imgs = np.vstack([imgs_pos, neg_list[sn_rand]])
except:
print(imgs_pos.shape)
print(neg_list[sn_rand].shape)
print(sn_rand)
imgs_24 = []
imgs_12 = []
for img in imgs:
im_24 = img.copy()
im_12 = img.copy()
im_24.resize((24, 24, 3))
im_12.resize((12, 12, 3))
imgs_24.append(im_24)
imgs_12.append(im_12)
imgs_24 = np.array(imgs_24)
imgs_12 = np.array(imgs_12)
labels = np.vstack([np.ones((20,1)), np.zeros((40,1))])
sn_shf = np.array(random.sample(range(60), 60))
imgs = imgs[sn_shf]
imgs_24 = imgs_24[sn_shf]
imgs_12 = imgs_12[sn_shf]
labels = labels[sn_shf]
if i%10000==0 and i!=0:
saver_48.save(sess, 'model/model_net_48', global_step=global_step, write_meta_graph=False)
if i%1==0:
feed_dict = {
net_output['imgs']: imgs,
net_output['labels']: labels,
net_output['imgs_24']: imgs_24,
net_output['labels_24']: labels,
net_output['imgs_12']: imgs_12,
net_output['labels_12']: labels,
net_output['keep_prob']: 1.0,
net_output['keep_prob_24']: 1.0,
net_output['keep_prob_12']: 1.0
}
cost, accuracy, recall, lr = sess.run(
[net_output['cost'], net_output['accuracy'], net_output['recall'], learning_rate],
feed_dict=feed_dict)
print("Step %d, cost: %f, acc: %f, recall: %f, lr: %f"%(i, cost, accuracy, recall, lr))
# print("target: ", target)
# print("pred: ", pred)
# train
feed_dict = {
net_output['imgs']: imgs,
net_output['labels']: labels,
net_output['imgs_24']: imgs_24,
net_output['labels_24']: labels,
net_output['imgs_12']: imgs_12,
net_output['labels_12']: labels,
net_output['keep_prob']: 0.8,
net_output['keep_prob_24']: 1.0,
net_output['keep_prob_12']: 1.0
}
sess.run(train_step, feed_dict=feed_dict)
coord.request_stop()
except Exception as e:
coord.request_stop(e)
finally:
print('Done training.')
saver_48.save(sess, 'model/model_net_48', global_step=global_step, write_meta_graph=False)
coord.request_stop()
coord.join(threads)
sess.close()
if __name__ == '__main__':
train()