获取 TypeError：为图像输入创建 TFRecords

Question

为图像输入创建TFrecords：如下

        char_ids_padded, char_ids_unpadded = encode_utf8_string(text)
        print("char_ids_padded:"+str(char_ids_padded))
        print("char_ids_unpadded:"+str(char_ids_unpadded))
        tf_example = tf.train.Example(features=tf.train.Features(feature={
            'image/format': _bytes_feature(b'png'),
            'image/encoded': _bytes_feature(image.tostring()),
            'image/class': _int64_feature(char_ids_padded),
            'image/unpadded_class': _int64_feature(char_ids_unpadded),
            'height': _int64_feature(image.shape[0]),
            'width': _int64_feature(image.shape[1]),
            'orig_width': _int64_feature(image.shape[1]/num_of_views),
            'image/text': _bytes_feature(text)
            }))


def _int64_feature(value):
  return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))

def _bytes_feature(value):
  return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))

char_ids_padded、char_ids_unpadded 的输出如下：

char_ids_padded:[47, 13, 16, 13, 16, 16, 16, 52, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

char_ids_unpadded:[47, 13, 16, 13, 16, 16, 16, 52]

注意：char_ids_padded 是 int 类型的列表格式，仍然在使用 tf.train.Features 映射时，得到错误为 TypeError: [47, 13, 16, 13, 16, 16, 16, 52, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] 的类型为“class 'list'”，但预期为以下之一：("class 'int'",)

Answer 1

您已经将一个列表传递给tf.train.Int64List，因此您不需要创建一个包含_int64_feature 参数的新列表。也就是说，你应该尝试改变

tf.train.Int64List(value=[value])

到

tf.train.Int64List(value=value)

在_int64_feature 函数中。

当我运行以下代码时，它可以工作：

def _int64_feature(value):
    return tf.train.Feature(int64_list=tf.train.Int64List(value=value))

char_ids_padded = [47, 13, 16, 13, 16, 16, 16, 52, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
tf_example = tf.train.Example(features=tf.train.Features(feature={
     'image/class': _int64_feature(char_ids_padded),
}))