This repository has been archived by the owner on Feb 22, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathnet3_4.py
184 lines (162 loc) · 8.81 KB
/
net3_4.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
#-*-coding:UTF-8-*-
#!/usr/bin/python
import tensorflow as tf
import numpy as np
from datetime import datetime
import math
import time
from create_input1 import read_and_decode
#data_in输入图片
data_in=tf.placeholder(tf.float32,[None,256,256,1])
#dropout层的数据保留百分比
keep_prob=tf.placeholder(tf.float32)
is_test=tf.placeholder(tf.bool)
def conv_op(input_op,name,kh,kw,n_out,dh,dw,p):
n_in=input_op.get_shape()[-1].value
with tf.name_scope(name) as scope:
kernel=tf.get_variable(scope+"w",shape=[kh,kw,n_in,n_out],dtype=tf.float32,initializer=tf.contrib.layers.xavier_initializer_conv2d())
#kernel=tf.Variable(tf.truncated_normal([kh,kw,n_in,n_out],stddev=0.1),trainable=True,name='w')
conv=tf.nn.conv2d(input_op,kernel,(1,dh,dw,1),padding='SAME')
n_out_conv=conv.get_shape()[-1].value
bias_init_val=tf.constant(0.0,shape=[n_out_conv],dtype=tf.float32)
#offset_init_val=tf.constant(0.0,shape=[n_out_conv],dtype=tf.float32)
#scale_init_val=tf.constant(1.0,shape=[n_out_conv],dtype=tf.float32)
#offset=tf.Variable(offset_init_val,trainable=True,name='offset')
biases=tf.Variable(bias_init_val,trainable=True,name='b')
z=tf.nn.bias_add(conv,biases)
#scale=tf.Variable(scale_init_val,trainable=True,name='scale')
#z=tf.nn.bias_add(conv,biases)
#exp_moving_avg = tf.train.ExponentialMovingAverage(0.998)
#bnepsilon = 1e-5
#mean, variance = tf.nn.moments(conv, [0])
#update_moving_averages = exp_moving_avg.apply([mean, variance])
#m = tf.cond(test, lambda: exp_moving_avg.average(mean), lambda: mean)
#v = tf.cond(test, lambda: exp_moving_avg.average(variance), lambda: variance)
#z = tf.nn.batch_normalization(conv, m, v, offset, scale, bnepsilon)
activation=tf.nn.relu(z,name=scope)
p+=[kernel,biases]
return activation
#全链接层
def fc_op(input_op,name,n_out,p):
n_in=input_op.get_shape()[-1].value
with tf.name_scope(name) as scope:
kernel=tf.get_variable(scope+"w",shape=[n_in,n_out],dtype=tf.float32,initializer=tf.contrib.layers.xavier_initializer_conv2d())
biases=tf.Variable(tf.constant(0.0,shape=[n_out],dtype=tf.float32),trainable=True,name='b')
activation=tf.nn.relu_layer(input_op,kernel,biases,name=scope)
#activation=tf.matmul(input_op,kernel)+biases
p+=[kernel,biases]
return activation
#max-pooling最大池化层
def mpool_op(input_op,name,kh,kw,dh,dw):
return tf.nn.max_pool(input_op,ksize=[1,kh,kw,1],strides=[1,dh,dw,1],padding='SAME',name=name)
def apool_op(input_op,name,kh,kw,dh,dw):
return tf.nn.avg_pool(input_op,ksize=[1,kh,kw,1],strides=[1,dh,dw,1],padding='SAME',name=name)
p=[]
#具体网络结构(改）
conv1_1=conv_op(data_in,name='conv1_1',kh=3,kw=3,n_out=64,dh=1,dw=1,p=p)
conv1_2=conv_op(conv1_1,name='conv1_2',kh=3,kw=3,n_out=64,dh=2,dw=2,p=p)
#conv1_1=conv_op(data_in,name='conv1_1',kh=7,kw=7,n_out=64,dh=2,dw=2,p=p)
#pool1=mpool_op(conv1_1,name='pool1',kh=2,kw=2,dw=2,dh=2)
conv2_1=conv_op(conv1_2,name='conv2_1',kh=3,kw=3,n_out=128,dh=1,dw=1,p=p)
conv2_2=conv_op(conv2_1,name='conv2_2',kh=3,kw=3,n_out=128,dh=2,dw=2,p=p)
#conv2_1=conv_op(conv1_1,name='conv2_1',kh=7,kw=7,n_out=128,dh=2,dw=2,p=p)
#pool2=mpool_op(conv2_1,name='pool2',kh=2,kw=2,dw=2,dh=2)
conv3_1=conv_op(conv2_2,name='conv3_1',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
conv3_2=conv_op(conv3_1,name='conv3_2',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
#conv3_3=conv_op(conv3_2,name='conv2_3',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
#conv3_1=conv_op(conv2_1,name='conv3_1',kh=7,kw=7,n_out=256,dh=1,dw=1,p=p)
#pool3=mpool_op(conv2_1,name='pool3',kh=2,kw=2,dw=1,dh=1)
conv4_1=conv_op(conv3_2,name='conv4_1',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
conv4_2=conv_op(conv4_1,name='conv4_2',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
#conv4_3=conv_op(conv4_2,name='conv4_3',kh=3,kw=3,n_out=256,dh=1,dw=1,p=p)
#conv4_1=conv_op(conv3_1,name='conv4_1',kh=7,kw=7,n_out=512,dh=1,dw=1,p=p)
#pool4=mpool_op(conv4_1,name='pool4',kh=2,kw=2,dw=1,dh=1)
conv5_1=conv_op(conv4_2,name='conv5_1',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
conv5_2=conv_op(conv5_1,name='conv5_2',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
#conv5_3=conv_op(conv5_2,name='conv5_3',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
#conv5_1=conv_op(conv4_1,name='conv5_1',kh=7,kw=7,n_out=512,dh=1,dw=1,p=p)
pool5=mpool_op(conv5_2,name='pool5',kh=2,kw=2,dw=1,dh=1)
"""
#conv6_1=conv_op(pool5,name='conv6_1',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
#conv6_2=conv_op(conv6_1,name='conv6_1',kh=1,kw=1,n_out=512,dh=1,dw=1,p=p)
#conv6_3=conv_op(conv6_2,name='conv6_3',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
conv6_1=conv_op(pool5,name='conv6_1',kh=7,kw=7,n_out=512,dh=1,dw=1,p=p)
pool6=mpool_op(conv6_1,name='pool6',kh=2,kw=2,dw=2,dh=2)
conv7_1=conv_op(pool6,name='conv7_1',kh=7,kw=7,n_out=512,dh=1,dw=1,p=p)
#conv7_2=conv_op(conv7_1,name='conv7_2',kh=3,kw=3,n_out=512,dh=1,dw=1,p=p)
pool7=mpool_op(conv7_1,name='pool7',kh=2,kw=2,dw=2,dh=2)
conv8_1=conv_op(pool7,name='conv8_1',kh=7,kw=7,n_out=768,dh=1,dw=1,p=p)
#conv8_2=conv_op(conv8_1,name='conv8_2',kh=3,kw=3,n_out=768,dh=1,dw=1,p=p)
pool8=mpool_op(conv8_1,name='pool8',kh=2,kw=2,dw=2,dh=2)
conv9_1=conv_op(pool8,name='conv9_1',kh=7,kw=7,n_out=768,dh=1,dw=1,p=p)
#conv9_2=conv_op(conv9_1,name='conv9_2',kh=1,kw=1,n_out=768,dh=1,dw=1,p=p)
#conv9_3=conv_op(conv9_2,name='conv9_3',kh=3,kw=3,n_out=768,dh=1,dw=1,p=p)
pool9=mpool_op(conv9_1,name='pool9',kh=2,kw=2,dw=2,dh=2)
conv10_1=conv_op(pool9,name='conv10_1',kh=7,kw=7,n_out=768,dh=1,dw=1,p=p)
pool10=mpool_op(conv10_1,name='pool10',kh=2,kw=2,dw=2,dh=2)
"""
shp=pool5.get_shape()
flattened_shape=shp[1].value*shp[2].value*shp[3].value
resh1=tf.reshape(pool5,[-1,flattened_shape],name='resh1')
fc6=fc_op(resh1,name='fc6',n_out=768,p=p)
fc6_drop=tf.nn.dropout(fc6,keep_prob,name='fc6_drop')
fc7=fc_op(fc6_drop,name='fc7',n_out=512,p=p)
fc7_drop=tf.nn.dropout(fc7,keep_prob,name='fc7_drop')
fc8=fc_op(fc7_drop,name='fc8',n_out=336,p=p)
fc8=tf.nn.softmax(fc8)
#values,indices=tf.nn.top_k(fc8,5)
#正确的数据标签
label=tf.placeholder(tf.float32,[None,336])
#计算准确率BP算法（后向传播)
cross_entropy=tf.reduce_mean(-tf.reduce_sum(label*tf.log(tf.clip_by_value(fc8,1e-10,1.0))))
#cross_entropy=tf.nn.softmax_cross_entropy_with_logits_v2(logits=fc8,labels=label)
#correct_label=tf.reshape(tf.argmax(label,1),[16,1])
#correct_prediction=tf.cast(tf.equal(tf.cast(indices,tf.int64),correct_label),tf.int32)
#correct_prediction=tf.reduce_sum(correct_prediction,reduction_indices=[1])
#correct_prediction=tf.reduce_sum(correct_prediction,reduction_indices=[0])
#accuracy=tf.cast(correct_prediction,tf.float32)/16
##
correct_prediction=tf.equal(tf.argmax(fc8,1),tf.argmax(label,1))
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
#梯度下降算法
train_step=tf.train.GradientDescentOptimizer(learning_rate=1e-4).minimize(cross_entropy)
#train_step=tf.train.AdamOptimizer(learning_rate=1e-4).minimize(cross_entropy)
#开始训练
train_img,train_label=read_and_decode("train_per.tfrecords")
test_img,test_label=read_and_decode("test_per.tfrecords")
train_img_batch,train_label_batch=tf.train.shuffle_batch([train_img,train_label],batch_size=128,capacity=2000,min_after_dequeue=1)
test_img_batch,test_label_batch=tf.train.shuffle_batch([test_img,test_label],batch_size=16,capacity=200,min_after_dequeue=1)
avg=0;
for_test,_=tf.nn.top_k(data_in,1)
init=tf.global_variables_initializer()
saver=tf.train.Saver()
with tf.Session() as sess:
sess.run(init)
coord=tf.train.Coordinator()
threads=tf.train.start_queue_runners(coord=coord)
try:
for i in range(2000):
train_imgs,train_labels=sess.run([train_img_batch,train_label_batch])
test_imgs,test_labels=sess.run([test_img_batch,test_label_batch])
test_labels=tf.one_hot(test_labels,336,dtype=tf.float32).eval()
train_labels=tf.one_hot(train_labels,336,dtype=tf.float32).eval()
train_imgs=np.array(train_imgs)
test_imgs=np.array(test_imgs)
#kernel=p[0]
#print(kernel.eval())
train_step.run({data_in:train_imgs,label:train_labels,keep_prob:1.0})
if i%10==0:
print(sess.run(cross_entropy,feed_dict={data_in:test_imgs,label:test_labels,keep_prob:0.8}))
#print(sess.run(cross_entropy,feed_dict={data_in:train_imgs,label:train_labels,keep_prob:0.9}))
acc=accuracy.eval({data_in:test_imgs,label:test_labels,keep_prob:1.0})
#print(accuracy.eval({data_in:test_imgs,label:test_labels,keep_prob:1.0}))
print(acc)
if i > 100:
avg=avg+acc
print('the total avg: ',avg/((i/10)-10),i)
saver.save(sess,'./model/model_smallk_keep/model.ckpt')
#except tf.errors.OpError:
#print('Oops,Errors!')
finally:
coord.request_stop()
coord.join(threads)