Example of model specifications to inputs and outputs.

TFGraph_Edit_Specification.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@author: jiankaiwang (https://jiankaiwang.no-ip.biz/)
@version:
  Tensorflow: 1.x (developed >= 1.13.2)
@description:
  Example of model specifications to inputs and outputs.
@dependency:
  OperateFrozenModel (TF1_FrozenModel.py, https://gist.github.com/jiankaiwang/24cc1bc8b38ce72bba73f7fb326f7b9e)
@changelog (main):
  2019-04: initial commit
  2019-09: released on gist.github.com
"""

import tensorflow as tf
import numpy as np
import OperateFrozenModel

# In[]

pb_path = "/Users/jiankaiwang/Desktop/output_graph.pb"
merged_pb_path = "/Users/jiankaiwang/Desktop/merged_graph.pb"
    
# In[]

tf.reset_default_graph()

# merged_graph 
merged_graph = tf.Graph()
with merged_graph.as_default():
  
  # defined a specific input
  # shape [None, 224, 224, 3] depends on your model's input
  inputs = tf.placeholder(tf.float32, [None, 224, 224, 3], "input")
  
  # import a trained model graph
  graph_def = tf.GraphDef()
  with tf.gfile.GFile(pb_path, "rb") as fin:
    graph_def.ParseFromString(fin.read())
          
  graph_outputs, = tf.import_graph_def(
      graph_def, 
      input_map={"Placeholder:0": inputs},
      return_elements=["final_result:0"], 
      name="")
  
  # defined a specific output
  outputs = tf.identity(graph_outputs, name="output")
  
  # exported as a frozen model
  with tf.Session() as sess:    
    state, graph = OperateFrozenModel.save_sess_into_frozen_model(
        sess, ["output"], merged_pb_path)
    print(state, graph)
  
  
# In[]
    
# sample input
sampled = np.random.randn(1, 224, 224, 3)

_, merged_graph = OperateFrozenModel.load_frozen_model(merged_pb_path)
with merged_graph.as_default():
  
  inputs = merged_graph.get_tensor_by_name("input:0")
  outputs = merged_graph.get_tensor_by_name("output:0")
  
  with tf.Session() as sess:
    print(sess.run(outputs, feed_dict={inputs: sampled}))