第一步 生成TFRecod Writer$1 import tensorflow as tf
$2 writer = tf.python_io.TFRecordWriter(path, options=None)
其中,path: 文件的存放路径,例如path='.../train.tfrecord'
options:TFRecordOptions对象,定义TFRecord文件保存的压缩格式,如下:# 三种文件压缩方式,默认方式3
# 方式1
writer = tf.python_io.TFRecordWriter(path,
options=tf.python_io.TFRecordOptions(tf.python_io.TFRecordCompressionType.ZLIB))
# 方式2
writer = tf.python_io.TFRecordWriter(path,
options=tf.python_io.TFRecordOptions(tf.python_io.TFRecordCompressionType.GZIP))
# 方式3
writer = tf.python_io.TFRecordWriter(path,
options=tf.python_io.TFRecordOptions(tf.python_io.TFRecordCompressionType.NONE))
第二步 tf.train.Feature 生成协议信息$1 feature = {"image": tf.train.Feature(bytes_list=tf.train.BytesList(value=[image.tobytes()])),