-
Notifications
You must be signed in to change notification settings - Fork 0
/
fingerprit.py
289 lines (247 loc) · 15 KB
/
fingerprit.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
"""
during training process:
input: 160*160
output: 20*20 = 400
total: *8 downsampling
"""
import tensorflow as tf
import numpy as np
from functools import reduce
class FingerPrint:
def __init__(self, vgg19_npy_path=None, trainable=True, dropout=0.5):
if vgg19_npy_path is not None:
self.data_dict = np.load(vgg19_npy_path, encoding='latin1').item()
else:
self.data_dict = None
# dictionary to store all the data of the network
self.var_dict = {}
# dictionary to store the value of all the variable
# format: self.var_dict[(name, idx)] = var
self.trainable = trainable
self.dropout = dropout
def build(self, image, batch_size, train_mode=True):
"""
network structure
"""
self.train_mode = train_mode
# assert image.shape[1:] == [160, 160, 1]
self.conv1_11 = self.conv_layer(image, 1, 64, "conv1_11", conv_size=7)
self.conv1_21 = self.conv_layer(self.conv1_11, 64, 128, "conv1_21", conv_size=7, stride=2)
self.relu11 = tf.nn.relu(self.conv1_21)
# self.bn1 = self.batch_norm_layer(self.relu1, name='bn1')
# self.pool1 = self.max_pool(self.relu1, 'pool1')
# self.conv1_12 = self.conv_layer(image, 1, 64, "conv1_12", conv_size=7)
# self.conv1_22 = self.conv_layer(self.conv1_12, 64, 128, "conv1_22", conv_size=7, stride=2)
# self.relu12 = tf.nn.relu(self.conv1_22)
# # self.bn1 = self.batch_norm_layer(self.relu1, name='bn1')
# # self.pool1 = self.max_pool(self.relu1, 'pool1')
# self.conv1_13 = self.conv_layer(image, 1, 64, "conv1_13", conv_size=7)
# self.conv1_23 = self.conv_layer(self.conv1_13, 64, 128, "conv1_23", conv_size=7, stride=2)
# self.relu13 = tf.nn.relu(self.conv1_23)
# # self.bn1 = self.batch_norm_layer(self.relu1, name='bn1')
# # self.pool1 = self.max_pool(self.relu1, 'pool1')
self.conv2_11 = self.conv_layer(self.relu11, 128, 256, "conv2_11", conv_size=5, stddev=0.001)
self.conv2_21 = self.conv_layer(self.conv2_11, 256, 256, "conv2_21", conv_size=5, stride=2)
self.relu21 = tf.nn.relu(self.conv2_21)
# self.bn2 = self.batch_norm_layer(self.relu2,name='bn2')
# self.pool2 = self.max_pool(self.relu2, 'pool2')
# self.conv2_12 = self.conv_layer(self.relu12, 128, 256, "conv2_12", conv_size=5, stddev=0.001)
# self.conv2_22 = self.conv_layer(self.conv2_12, 256, 256, "conv2_22", conv_size=5, stride=2)
# self.relu22 = tf.nn.relu(self.conv2_22)
# # self.bn2 = self.batch_norm_layer(self.relu2,name='bn2')
# # self.pool2 = self.max_pool(self.relu2, 'pool2')
# self.conv2_13 = self.conv_layer(self.relu13, 128, 256, "conv2_13", conv_size=5, stddev=0.001)
# self.conv2_23 = self.conv_layer(self.conv2_13, 256, 256, "conv2_23", conv_size=5, stride=2)
# self.relu23 = tf.nn.relu(self.conv2_23)
# self.bn2 = self.batch_norm_layer(self.relu2,name='bn2')
# self.pool2 = self.max_pool(self.relu2, 'pool2')
self.conv3_11 = self.conv_layer(self.relu21, 256, 512, "conv3_11", conv_size=3)
self.conv3_21 = self.conv_layer(self.conv3_11, 512, 512, "conv3_21", conv_size=3, stride=2) # 8-times downsampling
# self.conv3_3 = self.conv_layer(self.conv3_2, 256, 256, "conv3_3")
# self.conv3_4 = self.conv_layer(self.conv3_3, 256, 256, "conv3_4")
self.relu31 = tf.nn.relu(self.conv3_21)
# self.bn3 = self.batch_norm_layer(self.relu3, name='bn3')
# self.pool3 = self.max_pool(self.relu3, 'pool3')
# self.conv3_12 = self.conv_layer(self.relu22, 256, 512, "conv3_12", conv_size=3)
# self.conv3_22 = self.conv_layer(self.conv3_12, 512, 512, "conv3_22", conv_size=3, stride=2) # 8-times downsampling
# # self.conv3_3 = self.conv_layer(self.conv3_2, 256, 256, "conv3_3")
# # self.conv3_4 = self.conv_layer(self.conv3_3, 256, 256, "conv3_4")
# self.relu32 = tf.nn.relu(self.conv3_22)
# # self.bn3 = self.batch_norm_layer(self.relu3, name='bn3')
# # self.pool3 = self.max_pool(self.relu3, 'pool3')
# self.conv3_13 = self.conv_layer(self.relu23, 256, 512, "conv3_13", conv_size=3)
# self.conv3_23 = self.conv_layer(self.conv3_13, 512, 512, "conv3_23", conv_size=3, stride=2) # 8-times downsampling
# # self.conv3_3 = self.conv_layer(self.conv3_2, 256, 256, "conv3_3")
# # self.conv3_4 = self.conv_layer(self.conv3_3, 256, 256, "conv3_4")
# self.relu33 = tf.nn.relu(self.conv3_23)
# self.bn3 = self.batch_norm_layer(self.relu3, name='bn3')
# self.pool3 = self.max_pool(self.relu3, 'pool3')
# multiple layers rate=2,4,6.
self.conv4_11 = self.atrous_layer(self.relu31, 512, 256, "conv4_11", rate=1)
self.conv4_12 = self.conv_layer(self.conv4_11, 256, 256, "conv4_12")
self.relu411 = tf.nn.relu(self.conv4_12)
self.conv4_13 = self.atrous_layer(self.relu31, 512, 256, "conv4_13", rate=2)
self.conv4_14 = self.atrous_layer(self.conv4_13, 256, 256, "conv4_14", rate=2)
self.relu412 = tf.nn.relu(self.conv4_14)
# self.conv4_13 = self.atrous_layer(self.relu31, 512, 256, "conv4_13", rate=6)
# self.relu413 = tf.nn.relu(self.conv4_13)
# self.pool4 = self.max_pool(self.relu4, 'pool4')
self.conv5_1 = self.conv_layer(0.7 * self.relu411 + 0.3 * self.relu412, 256, 256, "conv5_1", conv_size=1)
# multiple layers rate=2,4,6.
self.conv4_21 = self.atrous_layer(self.relu31, 512, 256, "conv4_21", rate=1)
self.conv4_22 = self.conv_layer(self.conv4_21, 256, 256, "conv4_22")
self.relu421 = tf.nn.relu(self.conv4_22)
self.conv4_23 = self.atrous_layer(self.relu31, 512, 256, "conv4_23", rate=2)
self.conv4_24 = self.conv_layer(self.conv4_23, 256, 256, "conv4_24")
self.relu422 = tf.nn.relu(self.conv4_24)
# self.conv4_23 = self.atrous_layer(self.relu32, 512, 256, "conv4_23", rate=6)
# self.relu423 = tf.nn.relu(self.conv4_23)
# self.pool4 = self.max_pool(self.relu4, 'pool4')
self.conv5_2 = self.conv_layer(0.7 * self.relu421 + 0.3 * self.relu422, 256, 256, "conv5_2", conv_size=1)
# multiple layers rate=2,4,6.
self.conv4_31 = self.atrous_layer(self.relu31, 512, 256, "conv4_31", rate=1)
self.conv4_32 = self.conv_layer(self.conv4_31, 256, 256, "conv4_32")
self.relu431 = tf.nn.relu(self.conv4_32)
self.conv4_33 = self.atrous_layer(self.relu31, 512, 256, "conv4_33", rate=2)
self.conv4_34 = self.conv_layer(self.conv4_33, 256, 256, "conv4_32")
self.relu432 = tf.nn.relu(self.conv4_34)
# self.conv4_33 = self.atrous_layer(self.relu33, 512, 256, "conv4_33", rate=6)
# self.relu433 = tf.nn.relu(self.conv4_33)
# self.pool4 = self.max_pool(self.relu4, 'pool4')
self.conv5_3 = self.conv_layer(0.7 * self.relu431 + 0.3 * self.relu432, 256, 256, "conv5_3", conv_size=1)
self.fc81 = self.conv_layer(self.conv5_1, 256, 1, "fc8_1", conv_size=1)
self.fc82 = self.conv_layer(self.conv5_2, 256, 1, "fc8_2", conv_size=1)
self.fc83 = self.conv_layer(self.conv5_3, 256, 1, "fc8_3", conv_size=1)
self.fc9 = tf.concat(axis=3, values=[self.fc81, self.fc82, self.fc83]) # 三层输出，每层都在(0,0.1)
def getoutput(self, output): # 得到转换前的三层值，范围为[0, 179],并使用boosting得到最后的角度预测输出
change = tf.round(tf.multiply(output, 1790))
final = change
change1 = tf.nn.relu(change) # 负值变为0
change = 179 - tf.nn.relu(179 - change1) #大于179的值变为179
# 变换到(0,180)范围内
layer1, layer2, layer3 = tf.split(change, 3, -1)
# zeros = tf.zeros(tf.shape(layer1))
layer21 = tf.nn.relu(layer2 - 120) # 取出（120,179），变为(0,59)
layer22 = 119 - tf.nn.relu(119 - layer2) # 取出（0,119），其余变为119
temp2 = tf.cast(tf.equal(layer22, layer2), tf.float32) # 变为119的位置输出为0
layer2 = layer21 + (layer22 + 60) * temp2
layer31 = tf.nn.relu(layer3 - 60) # 取出（60,179），变为（0,119）
layer32 = 59 - tf.nn.relu(59 - layer3) # 取出（0,59），其余变为59
temp3 = tf.cast(tf.equal(layer32, layer3), tf.float32)
layer3 = layer31 + (layer32 + 120) * temp3
final = tf.concat(axis=3, values=[layer1, layer2, layer3])
# 得到boosting选择后的输出
layer4 = tf.abs(layer2 - layer1) # 查layer1和2的误差
layer41 = tf.nn.relu(layer4 - 15) + 15 # 小于x的输出全部变成x
temp4 = tf.cast(tf.equal(layer41, layer4), tf.float32) # 小于x的位置输出为0，大于x为1
final = tf.floor((layer1 + layer2) / 2) * (1 - temp4) + layer3 * temp4
return final
# def batch_norm_layer(self, x, name):
# with tf.variable_scope(name+'_bn'):
# beta = tf.Variable(tf.constant(0.0, shape=[x.shape[-1]]), name='beta', trainable=True)
# gamma = tf.Variable(tf.constant(1.0, shape=[x.shape[-1]]), name='gamma', trainable=True)
# axises = list(range(len(x.shape) - 1)) # [0 1 2]
# batch_mean, batch_var = tf.nn.moments(x, axises) # 得到输入层的mean和var
# ema = tf.train.ExponentialMovingAverage(decay=0.5) # 衰减系数0.5
# # print(self.train_mode)
#
# def mean_var_with_update():
# ema_apply_op = ema.apply([batch_mean, batch_var]) # 保持更新的变量组
# with tf.control_dependencies([ema_apply_op]): # 下文必须在ema_apply_op完成后完成
# return tf.identity(batch_mean), tf.identity(batch_var) # 返回的始终为更新后的值
#
# # mean, var = mean_var_with_update()
# mean, var = tf.cond(self.train_mode,
# lambda: mean_var_with_update(),
# lambda: (ema.average(batch_mean), ema.average(batch_var))
# )
# normed = tf.nn.batch_normalization(x, mean, var, beta, gamma, 1e-3)
# return normed
def avg_pool(self, bottom, name, stride=2):
return tf.nn.avg_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, stride, stride, 1], padding='SAME', name=name)
def max_pool(self, bottom, name, stride=2):
return tf.nn.max_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, stride, stride, 1], padding='SAME', name=name)
def conv_layer(self, bottom, in_channels, out_channels, name, padding='SAME', conv_size=3, stddev=0.001, stride=1):
with tf.variable_scope(name):
filt, conv_biases = self.get_conv_var(conv_size, in_channels, out_channels, name, stddev)
conv = tf.nn.conv2d(bottom, filt, [1, stride, stride, 1], padding=padding)
bias = tf.nn.bias_add(conv, conv_biases)
# relu = tf.nn.relu(bias)
return bias
def atrous_layer(self, bottom, in_channels, out_channels, name, padding='SAME', rate=2, stddev=0.001): # 空洞卷积
filt, conv_biases = self.get_conv_var(3, in_channels, out_channels, name, stddev)
conv = tf.nn.atrous_conv2d(bottom, filt, rate, padding=padding)
bias = tf.nn.bias_add(conv, conv_biases)
return bias
def fc_layer(self, bottom, in_size, out_size, name):
with tf.variable_scope(name):
weights, biases = self.get_fc_var(in_size, out_size, name)
x = tf.reshape(bottom, [-1, in_size])
fc = tf.nn.bias_add(tf.matmul(x, weights), biases)
return fc
def get_conv_var(self, filter_size, in_channels, out_channels, name, stddev): # name = 'convx_x'
initial_value = tf.truncated_normal([filter_size, filter_size, in_channels, out_channels], 0.0, stddev)
filters = self.get_var(initial_value, name, 0, name + "_filters")
initial_value = tf.truncated_normal([out_channels], .0, stddev) # shape, mean, stddev
biases = self.get_var(initial_value, name, 1, name + "_biases")
return filters, biases
def get_fc_var(self, in_size, out_size, name):
initial_value = tf.truncated_normal([in_size, out_size], 0.0, 0.001)
weights = self.get_var(initial_value, name, 0, name + "_weights")
initial_value = tf.truncated_normal([out_size], .0, .001)
biases = self.get_var(initial_value, name, 1, name + "_biases")
return weights, biases
def get_var(self, initial_value, name, idx, var_name):
if self.data_dict is not None and name in self.data_dict:
value = self.data_dict[name][idx]
else:
value = initial_value
print(name + ' value initial')
if self.trainable:
var = tf.Variable(value, name=var_name)
else:
var = tf.constant(value, dtype=tf.float32, name=var_name)
self.var_dict[(name, idx)] = var
# print var_name, var.get_shape().as_list()
# print(var.get_shape(), initial_value.get_shape())
assert var.get_shape() == initial_value.get_shape()
return var
def save_npy(self, sess, npy_path="./test2-save.npy"):
assert isinstance(sess, tf.Session)
data_dict = {}
for (name, idx), var in list(self.var_dict.items()):
# format of var_dict.items(): dict_items([((name1, idx1), var1), ((name2, idx2), var2)])
var_out = sess.run(var)
if name not in data_dict:
data_dict[name] = {}
data_dict[name][idx] = var_out
np.save(npy_path, data_dict)
print(("file saved", npy_path))
return npy_path
def get_var_count(self):
count = 0
for v in list(self.var_dict.values()):
count += reduce(lambda x, y: x * y, v.get_shape().as_list())
return count
#read tfrecord file
def read_and_decode(self, filename, epoch=None):
filename_queue = tf.train.string_input_producer\
([filename], num_epochs=epoch, shuffle=True)#生成一个random queue队列
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)#返回文件名和文件
features = tf.parse_single_example(serialized_example,
features={
'label': tf.FixedLenFeature([], tf.string),
'img_raw': tf.FixedLenFeature([], tf.string),
}) # 将image数据和label取出来
img0 = tf.decode_raw(features['img_raw'], tf.uint8)
img0 = tf.reshape(img0, [160, 160, 1]) # reshape为128*128的1通道图片
img0 = tf.cast(img0, tf.float32)
mean = tf.reduce_mean(img0)
std = tf.sqrt(tf.reduce_mean((img0-mean)**2))
img0 = (tf.cast(img0, tf.float32) - mean) * (1./std) # 白化
label = tf.decode_raw(features['label'], tf.float64) # 在流中抛出label张量
label = tf.cast(label, tf.float32)
label = tf.reshape(label, [20, 20, 3])
label = label * (1./1790) # 变换到（-0.1，0）
return img0, label