Tensorflow分批量读取数据

转载

转自:Tensorflow分批量读取数据

之前的博客里使用tf读取数据都是每次fetch一条记录,实际上大部分时候需要fetch到一个batch的小批量数据,在tf中这一操作的明显变化就是tensor的rank发生了变化,我目前使用的人脸数据集是灰度图像,因此大小是92*112的,所以最开始fetch拿到的图像数据集经过reshape之后就是一个rank为2的tensor,大小是92*112的(如果考虑通道,也可以reshape为rank为3的,即92*112*1)。如果加入batch,比如batch大小为5,那么拿到的tensor的rank就变成了3,大小为5*92*112。

下面规则化的写一下读取数据的一般流程,按照官网的实例,一般把读取数据拆分成两个大部分,一个是函数专门负责读取数据和解码数据,一个函数则负责生产batch。

import tensorflow as tf

def read_data(fileNameQue):

    reader = tf.TFRecordReader()
    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)
    img = tf.reshape(img, [92,112]) # 恢复图像原始大小
    label = tf.cast(features["label"], tf.int32)

    return img, label

def batch_input(filename, batchSize):

    fileNameQue = tf.train.string_input_producer([filename], shuffle=True)
    img, label = read_data(fileNameQue) # fetch图像和label
    min_after_dequeue = 1000
    capacity = min_after_dequeue+3*batchSize
    # 预取图像和label并随机打乱,组成batch,此时tensor rank发生了变化,多了一个batch大小的维度
    exampleBatch,labelBatch = tf.train.shuffle_batch([img, label],batch_size=batchSize, capacity=capacity,
                                                     min_after_dequeue=min_after_dequeue)
    return exampleBatch,labelBatch

if __name__ == "__main__":

    init = tf.initialize_all_variables()
    exampleBatch, labelBatch = batch_input("./data/faceTF.tfrecords", batchSize=10)

    with tf.Session() as sess:

        sess.run(init)
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(coord=coord)

        for i in range(100):
            example, label = sess.run([exampleBatch, labelBatch])
            print(example.shape)

        coord.request_stop()
        coord.join(threads)

 
 
  • 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

读取数据和解码数据与之前基本相同,针对不同格式数据集使用不同阅读器和解码器即可,后面是产生batch,核心是tf.train.shuffle_batch这个函数,它相当于一个蓄水池的功能,第一个参数代表蓄水池的入水口,也就是逐个读取到的记录,batch_size自然就是batch的大小了,capacity是蓄水池的容量,表示能容纳多少个样本,min_after_dequeue是指出队操作后还可以供随机采样出批量数据的样本池大小,显然,capacity要大于min_after_dequeue,官网推荐:min_after_dequeue + (num_threads + a small safety margin) * batch_size,还有一个参数就是num_threads,表示所用线程数目。

min_after_dequeue这个值越大,随机采样的效果越好,但是消耗的内存也越大。

  • 0
    点赞
  • 5
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值