Tensorflow服务，在线预测:如何构建一个接受'image_bytes‘作为输入张量名称的signature_def？

import tensorflow as tf
from tensorflow.python.saved_model.signature_constants import predict_signature_def

model = tf.saved_model.load('/path/to/model')

input_tensor_name = 'image_bytes'
input_tensor = tf.TensorSpec(shape=(None,), dtype=tf.string, name=input_tensor_name)

output_tensor_name = 'output'
output_tensor = tf.TensorSpec(shape=(None,), dtype=tf.float32, name=output_tensor_name)

signature_def = predict_signature_def(inputs={input_tensor_name: input_tensor}, outputs={output_tensor_name: output_tensor})

builder = tf.saved_model.Builder('/path/to/export_dir')
builder.add_meta_graph_and_variables(tf.compat.v1.get_default_session(), [tf.saved_model.tag_constants.SERVING], signature_def_map={tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: signature_def})
builder.save()