TF：利用TF读取数据操作，将CIFAR-10 数据集中的训练图片读取出来，并保存为.jpg格式

输出结果

核心代码

def inputs_origin(data_dir):

 filenames = [os.path.join(data_dir, 'data_batch_%d.bin' % i)

              for i in range(1, 6)]

……

 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__':

 with tf.Session() as sess:

   reshaped_image = inputs_origin('cifar10_data/cifar-10-batches-bin')

   threads = tf.train.start_queue_runners(sess=sess)

   sess.run(tf.global_variables_initializer())

   if not os.path.exists('cifar10_data/raw/'):

     os.makedirs('cifar10_data/raw/')

   for i in range(30):

     image_array = sess.run(reshaped_image)

     scipy.misc.toimage(image_array).save('cifar10_data/raw/%d.jpg' % i)