python - 导出tensorflow模型时为"Cannot infer num from shape"

Question

我正在尝试导出像这样的 tensorflow 模型

  feature_spec = { 'words': tf.FixedLenSequenceFeature([], tf.int64, allow_missing=True) }

  def serving_input_receiver_fn():
      """Build the serving inputs."""
      serialized_tf_example = tf.placeholder(dtype=tf.string,
                                             shape=[1],
                                             name='input_example_tensor')
      features = tf.parse_example(serialized_tf_example, feature_spec)
      receiver_tensors = {'words': serialized_tf_example}
      return tf.estimator.export.ServingInputReceiver(features, receiver_tensors)

  export_dir = classifier.export_savedmodel(export_dir_base=args.job_dir,
                                            serving_input_receiver_fn=serving_input_receiver_fn)

但我收到此错误

Cannot infer num from shape (1, ?, 128, 128)

我不知道？来自哪里，我猜它来自tf.parse_example。关于我在这里做错了什么有什么想法吗？

Answer 1

在不知道完整原因的情况下，这段代码似乎工作得很好

  def serving_input_receiver_fn():
    feature_spec = { "words": tf.FixedLenFeature(dtype=tf.int64, shape=[4]) }
    return tf.estimator.export.build_parsing_serving_input_receiver_fn(feature_spec)()