# -*- coding: utf-8 -*-
"""
Created on Tue Aug 7 20:45:01 2018
@author: lenovo
"""
import cifar10_input
import tensorflow as tf
import os
import scipy.misc
def inputs_origin(data_dir):
filenames = [os.path.join(data_dir, 'data_batch_%d.bin' % i)
for i in range(1, 6)]
# 判断文件是否存在
for f in filenames:
if not tf.gfile.Exists(f):
raise ValueError('Failed to find file: ' + f)
# 将文件名的list包装成TensorFlow中queue的形式
filename_queue = tf.train.string_input_producer(filenames)
read_input = cifar10_input.read_cifar10(filename_queue)
reshaped_image = tf.cast(read_input.uint8image, tf.float32)
return reshaped_image
if __name__ == '__main__':
# 创建一个会话sess
with tf.Session() as sess:
# 调用inputs_origin。cifar10_data/cifar-10-batches-bin是我们下载的数据的文件夹位置
reshaped_image = inputs_origin('C:\\Users\\lenovo\\Desktop\\cifar10_data\\cifar-10-batches-bin')
# 这一步start_queue_runner很重要。
# 我们之前有filename_queue = tf.train.string_input_producer(filenames)
# 这个queue必须通过start_queue_runners才能启动
# 缺少start_queue_runners程序将不能执行
threads = tf.train.start_queue_runners(sess=sess)
# 变量初始化
sess.run(tf.global_variables_initializer())
# 创建文件夹cifar10_data/raw/
if not os.path.exists('cifar10_data/raw/'):
os.makedirs('cifar10_data/raw/')
# 保存30张图片
for i in range(30):
# 每次sess.run(reshaped_image)，都会取出一张图片
image_array = sess.run(reshaped_image)
# 将图片保存
scipy.misc.toimage(image_array).save('cifar10_data/raw/%d.jpg' % i)