tensorflow tfrecord文件存儲

 

 

import tensorflow as tf import numpy as np import skimage from skimage import data, io, color from PIL import Image path = "1.tfrecords" img_path = '/data/test/img/1.png' with tf.python_io.TFRecordWriter(path) as writer: # list: int or float
    a = 1024 b = 10.24 c = [0.1, 0.2, 0.3] c = np.array(c).astype(np.float32).tobytes() d = [[1, 2], [3, 4]] d = np.array(d).astype(np.int8).tobytes() e = "Python" e = bytes(e, encoding='utf-8') img = io.imread(img_path) img = img.astype(np.uint8).tobytes() img2 = Image.open(img_path) img2 = img2.resize((256, 256)) img2 = img2.tobytes() example = tf.train.Example(features=tf.train.Features(feature={ 'a': tf.train.Feature(int64_list=tf.train.Int64List(value=[a])), 'b': tf.train.Feature(float_list=tf.train.FloatList(value=[b])), 'c': tf.train.Feature(bytes_list=tf.train.BytesList(value=[c])), 'd': tf.train.Feature(bytes_list=tf.train.BytesList(value=[d])), 'e': tf.train.Feature(bytes_list=tf.train.BytesList(value=[e])), 'image': tf.train.Feature(bytes_list=tf.train.BytesList(value=[img])), 'image2': tf.train.Feature(bytes_list=tf.train.BytesList(value=[img2])), })) writer.write(example.SerializeToString()) # 讀取
filename_queue = tf.train.string_input_producer([path]) _, serialized_example = tf.TFRecordReader().read(filename_queue) features = tf.parse_single_example(serialized_example, features={ 'a': tf.FixedLenFeature([], tf.int64), 'b': tf.FixedLenFeature([], tf.float32), 'c': tf.FixedLenFeature([], tf.string), 'd': tf.FixedLenFeature([], tf.string), 'e': tf.FixedLenFeature([], tf.string), 'image': tf.FixedLenFeature([], tf.string), 'image2': tf.FixedLenFeature([], tf.string), }) a = features['a'] # 返回是張量
b = features['b'] c = features['c'] c = tf.decode_raw(c, tf.float32) d = features['d'] d = tf.decode_raw(d, tf.int8) d = tf.reshape(d, [2, 2]) e = features['e'] img = tf.decode_raw(features['image'], tf.uint8) img = tf.reshape(img, shape=[256, 256, 3]) img2 = tf.decode_raw(features['image2'], tf.uint8) img2 = tf.reshape(img2, [256, 256,3]) with tf.Session() as sess: sess.run(tf.initialize_all_variables()) tf.train.start_queue_runners(sess=sess) print(sess.run([a, b, c, d, e])) e = sess.run(e) print(type(e), bytes.decode(e)) img = sess.run(img) io.imshow(img) img2 = sess.run(img2) io.imshow(img2)