Tenforflow之构建自己的cpkt训练模型

亦凉 2022-07-18 01:05 165阅读 0赞

学习谷歌的深度学习终于有点眉目了，给大家分享我的Tensorflow学习历程。

tensorflow的官方中文文档比较生涩，数据集一直采用的MNIST二进制数据集。并没有过多讲述怎么构建自己的图片数据集tfrecords。

先贴我的转化代码将图片文件夹下的图片转存tfrecords的数据集。

\[python\] view plain copy

###### \#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\# ######

# !/usr/bin/python2.7 #

# \-*- coding: utf-8 -*\- #

# Author : zhaoqinghui #

# Date : 2016.5.10 #

# Function: image convert to tfrecords #

###### \#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\# ######

import tensorflow as tf  
import numpy as np  
import cv2  
import os  
import os.path  
from PIL import Image

# 参数设置 #

###### \#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\# ######

train\_file = ‘train.txt’ \#训练图片  
name=’train’ \#生成train.tfrecords  
output\_directory=’./tfrecords’  
resize\_height=32 \#存储图片高度  
resize\_width=32 \#存储图片宽度

###### \#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\#\# ######

def \_int64\_feature(value):  
return tf.train.Feature(int64\_list=tf.train.Int64List(value=\[value\]))

def \_bytes\_feature(value):  
return tf.train.Feature(bytes\_list=tf.train.BytesList(value=\[value\]))

def load\_file(examples\_list\_file):  
lines = np.genfromtxt(examples\_list\_file, delimiter=” “, dtype=\[(‘col1’, ‘S120’), (‘col2’, ‘i8’)\])  
examples = \[\]  
labels = \[\]  
for example, label in lines:  
examples.append(example)  
labels.append(label)  
return np.asarray(examples), np.asarray(labels), len(lines)

def extract\_image(filename, resize\_height, resize\_width):  
image = cv2.imread(filename)  
image = cv2.resize(image, (resize\_height, resize\_width))  
b,g,r = cv2.split(image)  
rgb\_image = cv2.merge(\[r,g,b\])  
return rgb\_image

def transform2tfrecord(train\_file, name, output\_directory, resize\_height, resize\_width):  
if not os.path.exists(output\_directory) or os.path.isfile(output\_directory):  
os.makedirs(output\_directory)  
\_examples, \_labels, examples\_num = load\_file(train\_file)  
filename = output\_directory + “/” + name + ‘.tfrecords’  
writer = tf.python\_io.TFRecordWriter(filename)  
for i, \[example, label\] in enumerate(zip(\_examples, \_labels)):  
print(‘No.%d’ % (i))  
image = extract\_image(example, resize\_height, resize\_width)  
print(‘shape: %d, %d, %d, label: %d’ % (image.shape\[0\], image.shape\[1\], image.shape\[2\], label))  
image\_raw = image.tostring()  
example = tf.train.Example(features=tf.train.Features(feature=\{  
‘image\_raw’: \_bytes\_feature(image\_raw),  
‘height’: \_int64\_feature(image.shape\[0\]),  
‘width’: \_int64\_feature(image.shape\[1\]),  
‘depth’: \_int64\_feature(image.shape\[2\]),  
‘label’: \_int64\_feature(label)  
\}))  
writer.write(example.SerializeToString())  
writer.close()

def disp\_tfrecords(tfrecord\_list\_file):  
filename\_queue = tf.train.string\_input\_producer(\[tfrecord\_list\_file\])  
reader = tf.TFRecordReader()  
\_, serialized\_example = reader.read(filename\_queue)  
features = tf.parse\_single\_example(  
serialized\_example,  
features=\{  
‘image\_raw’: tf.FixedLenFeature(\[\], tf.string),  
‘height’: tf.FixedLenFeature(\[\], tf.int64),  
‘width’: tf.FixedLenFeature(\[\], tf.int64),  
‘depth’: tf.FixedLenFeature(\[\], tf.int64),  
‘label’: tf.FixedLenFeature(\[\], tf.int64)  
\}  
)  
image = tf.decode\_raw(features\[‘image\_raw’\], tf.uint8)  
\#print(repr(image))  
height = features\[‘height’\]  
width = features\[‘width’\]  
depth = features\[‘depth’\]  
label = tf.cast(features\[‘label’\], tf.int32)  
init\_op = tf.initialize\_all\_variables()  
resultImg=\[\]  
resultLabel=\[\]  
with tf.Session() as sess:  
sess.run(init\_op)  
coord = tf.train.Coordinator()  
threads = tf.train.start\_queue\_runners(sess=sess, coord=coord)  
for i in range(21):  
image\_eval = image.eval()  
resultLabel.append(label.eval())  
image\_eval\_reshape = image\_eval.reshape(\[height.eval(), width.eval(), depth.eval()\])  
resultImg.append(image\_eval\_reshape)  
pilimg = Image.fromarray(np.asarray(image\_eval\_reshape))  
pilimg.show()  
coord.request\_stop()  
coord.join(threads)  
sess.close()  
return resultImg,resultLabel

def read\_tfrecord(filename\_queuetemp):  
filename\_queue = tf.train.string\_input\_producer(\[filename\_queuetemp\])  
reader = tf.TFRecordReader()  
\_, serialized\_example = reader.read(filename\_queue)  
features = tf.parse\_single\_example(  
serialized\_example,  
features=\{  
‘image\_raw’: tf.FixedLenFeature(\[\], tf.string),  
‘width’: tf.FixedLenFeature(\[\], tf.int64),  
‘depth’: tf.FixedLenFeature(\[\], tf.int64),  
‘label’: tf.FixedLenFeature(\[\], tf.int64)  
\}  
)  
image = tf.decode\_raw(features\[‘image\_raw’\], tf.uint8)  
\# image  
tf.reshape(image, \[256, 256, 3\])  
\# normalize  
image = tf.cast(image, tf.float32) \* (1. /255) - 0.5  
\# label  
label = tf.cast(features\[‘label’\], tf.int32)  
return image, label

def test():  
transform2tfrecord(train\_file, name , output\_directory, resize\_height, resize\_width) \#转化函数  
img,label=disp\_tfrecords(output\_directory+’/’+name+’.tfrecords’) \#显示函数  
img,label=read\_tfrecord(output\_directory+’/’+name+’.tfrecords’) \#读取函数  
print label

if **name** == ‘**main**‘:  
test()

这样就可以得到自己专属的数据集.tfrecords了 ，它可以直接用于tensorflow的数据集。