猫狗大战 从 record 文件中读取一张图片并显示

原文链接: 猫狗大战 从 record 文件中读取一张图片并显示

上一篇: 猫狗大战数据集 record 形式写入与读取

下一篇: tensorflow flowers数据集 record文件 提取图像

record文件不会一次性全部加载进内存

写入的时候是按照rgb顺序写入的，使用opencv显示的时候需要做一次颜色转化

每次的图片显示的都是不一样的

import tensorflow as tf
import cv2 as cv

RECORD_PATH = "d:/data/cat_vs_dog/cat_vs_dog.record"


def read(path):
    queue = tf.train.string_input_producer([path])
    reader = tf.TFRecordReader()
    _, data = reader.read(queue)
    features = tf.parse_single_example(
        data, {
            'label': tf.FixedLenFeature([], tf.int64),
            'image': tf.FixedLenFeature([], tf.string)
        }
    )
    label = features['label']
    image = features['image']

    # 不知为何会出错...
    # image = tf.image.decode_image(image, channels=3)

    # 必须将图片转化为大小相同的矩阵，不然没办法进行批次训练
    image = tf.decode_raw(image, tf.uint8)
    image = tf.reshape(image, (224, 224, 3))
    # image  (224, 224, 3)
    print('image ', image.shape)

    image_batch, label_batch = tf.train.shuffle_batch(
        [image, label],
        batch_size=16,
        capacity=256,
        min_after_dequeue=128,
        num_threads=4
    )

    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())

        coord = tf.train.Coordinator()
        thread = tf.train.start_queue_runners(sess, coord)

        image_data, label_data = sess.run(
            [image_batch, label_batch]
        )

        # (16, 224, 224, 3) (16,)
        print(image_data.shape, label_data.shape)
        img = image_data[0]
        img = cv.cvtColor(img, cv.COLOR_BGR2RGB)

        # (224, 224, 3) 0
        print(img.shape, label_data[0])
        cv.imshow('image', img)
        cv.waitKey(0)
        coord.request_stop()
        # 超时时间30s
        coord.join(thread, 30)


if __name__ == '__main__':
    read(RECORD_PATH)