深度学习：TensorFlow笔记——制作Cifar数据集

      学习完Mnist手写体识别后，想用Cifar数据集制作成方便深度神经网络训练的二进制数据集文件格式。Cifar数据集包含 airplane、automobile、bird、cat、deer、dog、frog、horse、ship、truck10种分类，共6万张图片，其中训练集5万张，测试集1万张。一般情况下，训练的数据都是图片，如果直接读取图片进行训练，程序的执行效率不高。采用tfrecords可以把图片和对应的标签放在一个二进制文件中。接下来将介绍如何生成便于训练的Cifar数据集。具体思路是：从Cifar这个文件夹中遍历10个种类 的文件夹，然后根据每个文件夹的路径遍历所有的文件。

import tensorflow as tf
from PIL import Image
import os
image_train_path='./train/'
label_train_path='./test_label.txt'
tfRecord_train='./Cifar_train.tfrecords'
image_test_path='./test/'
label_test_path='./test_label.txt'
tfRecord_test='./Cifar_test.tfrecords'
data_path='./data'
def write_tfRecord(tfRecordName, image_path, label_path):
       writer = tf.python_io.TFRecordWriter(tfRecordName)
       num_pic = 0
       label_cnt = 0
       获取各个子文件夹的文件名
       Data_dir =  os.listdir(image_path)
       对获取的文件名进行排序，方便制作相应的标签
       Data_dir = sorted(Data_dir)
       for child_dir in Data_dir:
       获取每个子文件夹的路径
           img_path = image_path + child_dir +'/'
           pic_name = os.listdir(img_path)
           标签初始化
           labels = [0]*10
           labels[label_cnt] =1
           label_cnt +=1 
           循环遍历每个子文件夹里面的图片              
           for pic_path in pic_name:
               获取每个文件的路径，然后打开图片
               flie_path = img_path + pic_path
               img = Image.open(flie_path)
               将图片信息转换为二进制信息            
               img_raw = img.tobytes()
               把图片和对应的标签存入一个二进制文件
               example = tf.train.Example(features=tf.train.Features(feature={
                   'img_raw': tf.train.Feature(bytes_list=tf.train.BytesList(value=[img_raw])),
                   'label': tf.train.Feature(int64_list=tf.train.Int64List(value=labels))
                                            }))
            writer.write(example.SerializeToString())
            num_pic += 1
         writer.close()
         print("write tfrecord successful")
def generate_tfRecord():
    isExists = os.path.exists(data_path)
    if not isExists:
        os.makedirs(data_path)
        print 'the directory was created successfully'
    else:
        print 'dirctory already exists'
    write_tfRecord(tfRecord_train, image_train_path, label_train_path)
    write_tfRecord(tfRecord_test, image_test_path, label_test_path)

以上代码是写入数据函数，接下来是在训练的时候，如何读取训练数据。接着上面的程序继续

def read_tfRecord(tfRecord_path):
       filename_queue = tf.train.string_input_producer([tfRecord_path], shuffle=True)
       reader = tf.TFRecordReader()
       _, serialized_example = reader.read(filename_queue)
       features = tf.parse_single_example(serialized_example,
                                          features={
                                         'label': tf.FixedLenFeature([10], tf.int64),
                                          'img_raw': tf.FixedLenFeature([], tf.string)
                                          })

      img = tf.decode_raw(features['img_raw'], tf.uint8)
      img = tf.reshape(img,[32, 32,3])
      img = tf.cast(img, tf.float32) * (1. / 255)
      label = tf.cast(features['label'], tf.float32)
      return img, label
def get_tfrecord(num, isTrain=True):
    if isTrain:
        tfRecord_path = tfRecord_train
    else:
        tfRecord_path = tfRecord_test
    img, label = read_tfRecord(tfRecord_path)
    img_batch, label_batch = tf.train.shuffle_batch([img, label],
                                            batch_size  =num,
                                            num_threads =2,
                                            capacity = 2000,
                                            min_after_dequeue = 1000)
return img_batch, label_batch

def main():
    generate_tfRecord()

if __name__ == '__main__':
    main() 

运行这个上述代码即可得到训练集文件和测试集文件。