-
Notifications
You must be signed in to change notification settings - Fork 10
/
read.py
168 lines (127 loc) · 5.56 KB
/
read.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
# coding:utf-8
import os
import numpy as np
import tensorflow as tf
import transform
# 打开tensorflow的可视化工具
# tensorboard --logdir "/Users/yanyuanchi/code/python/readtffile/see"
# 参数存储路径
params_path = "/Users/yanyuanchi/code/python/readtffile/params/"
def save_np_array(array, path):
print "参数:", path
print "数据shape:", np.shape(array)
print "-------------------------------------------------------------------"
f = file(path, "wb")
array.flatten().astype("float32").tofile(f)
f.close()
# tf 的卷积的权值存储为(h, w, input_channel, output_channel)
# 需要转换成(output_channel, h, w, input_channel)
def save_tf_conv_np_array(array, path):
array = np.moveaxis(array, -1, 0)
save_np_array(array, path)
# tf 的卷积的权值存储为(h, w, output_channel, input_channel)
# 需要转换成(output_channel, h, w, input_channel)
def save_tf_tranpose_conv_np_array(array, path):
array = np.moveaxis(array, 2, 0)
outChannel, h, w, inChannel = np.shape(array)
for out in range(outChannel):
temp = array[out].copy()
for y in range(h):
for x in range(w):
in1 = temp[h - y - 1][w - x - 1]
out1 = array[out][y][x]
for l in range(inChannel):
out1[l] = in1[l]
save_np_array(array, path)
image_height = 228
image_width = 228
checkpoint_dir = "/Users/yanyuanchi/code/python/readtffile/model/la_muse.ckpt"
g = tf.Graph()
# allow_soft_placement=True : 如果你指定的设备不存在,允许TF自动分配设备
soft_config = tf.ConfigProto(allow_soft_placement=True)
soft_config.gpu_options.allow_growth = True
with g.as_default(), g.device("/cpu:0"), tf.Session(config=soft_config) as sess:
batch_shape = (1, image_height, image_width, 3)
img_placeholder = tf.placeholder(tf.float32, shape=batch_shape, name='img_placeholder')
preds = transform.net(img_placeholder)
saver = tf.train.Saver()
saver.restore(sess, checkpoint_dir)
variables = tf.trainable_variables()
# variables = tf.model_variables()
# for var in variables:
# print var.name
# print variables
# for var in variables:
# name = var.name
# realValue = var._variable.eval()
#
# print name
# print np.shape(realValue)
# print "-------------------------------------"
# 这个项目对应的结构是
# 3个卷积
# 1:卷积(weight),batchnormalization(alpha, beta) 3个变量
# 2:卷积(weight),batchnormalization(alpha, beta) 3个变量
# 3:卷积(weight),batchnormalization(alpha, beta) 3个变量
# 5个res层 bn 的变量顺序为 beta alpha
# 4:res层 卷积(weights),batchnormalization(alpha,beta)卷积(weights),batchnormalization(alpha,beta) 6个变量
# 5:res层 卷积(weights),batchnormalization(alpha,beta)卷积(weights),batchnormalization(alpha,beta) 6个变量
# 6:res层 卷积(weights),batchnormalization(alpha,beta)卷积(weights),batchnormalization(alpha,beta) 6个变量
# 7:res层 卷积(weights),batchnormalization(alpha,beta)卷积(weights),batchnormalization(alpha,beta) 6个变量
# 8:res层 卷积(weights),batchnormalization(alpha,beta)卷积(weights),batchnormalization(alpha,beta) 6个变量
# 两个转置卷积
# 9:转置卷积(weights),batchnormalization(alpha, beta) 3个变量
# 10:转置卷积(weights),batchnormalization(alpha, beta) 3个变量
# 卷积
# 11:卷积(weights),batchnormalization(alpha, beta) 3个变量
i = 0
for var in variables:
name = var.name
realValue = var._variable.eval()
if i < 9:
name = "conv" + str(i / 3 + 1)
if 0 == i % 3:
save_tf_conv_np_array(realValue, params_path + name + "_weight")
elif 1 == i % 3:
save_np_array(realValue, params_path + name + "_beta")
else:
save_np_array(realValue, params_path + name + "_alpha")
elif i < 39:
# res层
j = i - 9
name = "res" + str(j / 6 + 1)
if j % 6 < 3:
name += "_conv1"
else:
name += "_conv2"
z = (j % 6)
if 0 == z % 3:
save_tf_conv_np_array(realValue, params_path + name + "_weight")
elif 1 == z % 3:
save_np_array(realValue, params_path + name + "_beta")
else:
save_np_array(realValue, params_path + name + "_alpha")
elif i < 45:
j = i - 39
name = "transpose_conv" + str(j / 3 + 1)
z = j % 6
if 0 == z % 3:
# (h, w, outputchannel, inputchannel)
# to (outputchannel, h, w, inputchannel)
save_tf_tranpose_conv_np_array(realValue, params_path + name + "_weight")
elif 1 == z % 3:
save_np_array(realValue, params_path + name + "_beta")
else:
save_np_array(realValue, params_path + name + "_alpha")
elif i < 48:
j = i - 45
name = "conv4"
if 0 == j % 3:
save_tf_conv_np_array(realValue, params_path + name + "_weight")
elif 1 == j % 3:
save_np_array(realValue, params_path + name + "_beta")
else:
save_np_array(realValue, params_path + name + "_alpha")
else:
raise Exception("error")
i += 1