TensorFlow读取tfrecords数据

因为要用到TensorFlow，自然少不了数据的读取，这里我自己写了一个tfrecords的数据的读取函数

"""
Created on Wed Jun 28 13:56:35 2017

@author:liao
"""
import tensorflow as tf
import numpy as np
from PIL import Image
import pprint as pp
import os

#cwd = os.getcwd()
#file_path=[cwd+os.sep+'emotion-tfrecord'+os.sep+'emotion_image_data.tfrecords']
#fileNameQue = tf.train.string_input_producer(file_path)#从文件名列表，生成文件名队列
#reader = tf.TFRecordReader()#用tfrecords专用的reader去读
#key,value = reader.read(fileNameQue)
#features = tf.parse_single_example(value,features={ 'label': tf.FixedLenFeature([], tf.int64),
#                                           'img' : tf.FixedLenFeature([], tf.string),})
#
#img = tf.decode_raw(features["img"], tf.uint8)
#image_raw=tf.reshape(img,[48,48])#如果后面要接卷积神经网络的话，就化成这样的形式
#labels = tf.cast(features["label"], tf.int32)
#
##使用样本队列，让本队列自动打包，发送
#num_preprocess_threads = 16
#image, label = tf.train.shuffle_batch([image_raw,labels], batch_size = batch_size, num_threads=num_preprocess_threads, min_after_dequeue = 1, capacity = 10)
#
#
#init = tf.global_variables_initializer()#注意这里一定要有这初始化


#with tf.Session() as sess:
#
#    sess.run(init)
#    #线程协调器，控制打包读取的线程
#    coord = tf.train.Coordinator()
#    #让读取队列开始工作
#    threads = tf.train.start_queue_runners(coord=coord)
#    #从队列读取图像和标签
#    images,labels=sess.run([image,label])
#    print(images,labels)
#
#
#    coord.request_stop()
#    coord.join(threads)
def read_tfrecords(file_path,batch_size):
    fileNameQue = tf.train.string_input_producer([file_path])
    reader = tf.TFRecordReader()#用tfrecords专用的reader去读
    key,value = reader.read(fileNameQue)    
    features = tf.parse_single_example(value,features={ 'label': tf.FixedLenFeature([], tf.int64),
                                           'img' : tf.FixedLenFeature([], tf.string),})

    img = tf.decode_raw(features["img"], tf.uint8)
    image_raw=tf.reshape(img,[1,48,48])#如果后面要接卷积神经网络的话，就化成这样的形式[depth,height,width]
    image_raw = tf.transpose(image_raw, (1,2,0))# 
    image_raw = tf.cast(image_raw, tf.float32)
#    归一化操作
    image_raw= tf.image.per_image_standardization(image_raw)
#    
    labels = tf.cast(features["label"], tf.int32)
    #使用样本队列，让本队列自动打包，发送
    num_preprocess_threads = 16
    image, label = tf.train.shuffle_batch([image_raw,labels], batch_size = batch_size, num_threads=num_preprocess_threads, min_after_dequeue = 1, capacity = 1024)
    return image, tf.reshape(label, [batch_size])

屏蔽掉的部分是测试程序。有关数TensorFlow的数据的读取过程可以参考 点击打开链接