Tensorflow - optimize_for_inference_lib.optimize_for_inference丢失了tf.identity指定的输出节点

Question

TensorFlow版本1.1.0

我通过tf.identity指定“输入”、“输出”作为我的模型的输入和输出节点。

冻结后，我可以在graph_def.node中找到“输入”、“输出”。

优化后，“输入”仍然存在，但“输出”却丢失了。

`Found in frozen Traceback (most recent call last): File "/home/xiusir/.virtualenvs/ir/lib/python3.5/site-packages/tensorflow/python/framework/importer.py", line 480, in import_graph_def ret.append(name_to_op[name]) KeyError: 'output' During handling of the above exception, another exception occurred: Traceback (most recent call last): File "graph.py", line 43, in <module> output_tensor = tf.import_graph_def(graph_def, name='', return_elements=[MODEL_OUTPUT_TENSOR_NAME]) File "/home/xiusir/.virtualenvs/ir/lib/python3.5/site-packages/tensorflow/python/framework/importer.py", line 483, in import_graph_def 'Requested return_element %r not found in graph_def.' % name) ValueError: Requested return_element 'output' not found in graph_def.`

​

train.py

def inference(images):
  images = tf.identity(images, 'input')
  ... ...
  softmax_linear = tf.identity(softmax_linear, 'output')
  return softmax_linear

package.py --生成模型文件

# -*- coding: utf-8 -*-
# Preparing a TF model for usage in Android
# pylint: disable=missing-docstring
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import sys
import tensorflow as tf
from tensorflow.python.tools import freeze_graph
from tensorflow.python.tools import optimize_for_inference_lib

MODEL_NAME = 'carc34'
path='.'
input_graph_path = '%s/graph.pbtxt' % path
checkpoint_path = '%s/model.ckpt-100000' % path
input_saver_def_path = ""
input_binary = False
input_node_names = "input"
output_node_names = "output,softmax_linear/softmax_linear"
restore_op_name = "save/restore_all"
filename_tensor_name = "save/Const:0"
output_frozen_graph_name = 'frozen_'+MODEL_NAME+'.pb'
output_optimized_graph_name = 'optimized_'+MODEL_NAME+'.pb'
clear_devices = True

# Freeze the graph
freeze_graph.freeze_graph(input_graph_path, input_saver_def_path,
                      input_binary, checkpoint_path, output_node_names,
                      restore_op_name, filename_tensor_name,
                      output_frozen_graph_name, clear_devices, "")

# Optimize for inference
input_graph_def = tf.GraphDef()
with tf.gfile.Open(output_frozen_graph_name, "rb") as f:
    data = f.read()
    input_graph_def.ParseFromString(data)
    output_graph_def = optimize_for_inference_lib.optimize_for_inference(
            input_graph_def,
            input_node_names.split(","),  # an array of the input node(s)
            output_node_names.split(","), # an array of the output nodes
            tf.float32.as_datatype_enum)

# Save the optimized graph
f = tf.gfile.FastGFile(output_optimized_graph_name, "w")
f.write(output_graph_def.SerializeToString())

show_nodes.py -打印冻结和优化中的所有节点

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os.path
import numpy as np
import tensorflow as tf
from tensorflow.python.framework import graph_util
from tensorflow.python.framework import tensor_shape
from tensorflow.python.platform import gfile
from tensorflow.python.util import compat

MODEL_OUTPUT_TENSOR_NAME = 'output'
with tf.Graph().as_default() as graph:
  model_filename = os.path.join('.', 'frozen_carc34.pb')
  with gfile.FastGFile(model_filename, 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())
    ##for n in graph_def.node:
    ##  print (n.name)
    output_tensor = tf.import_graph_def(graph_def, name='', return_elements=[MODEL_OUTPUT_TENSOR_NAME])
    print ("Found in frozen")

  model_filename = os.path.join('.', 'optimized_carc34.pb')
  with gfile.FastGFile(model_filename, 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())
    ##for n in graph_def.node:
    ##  print (n.name)
    output_tensor = tf.import_graph_def(graph_def, name='', return_elements=[MODEL_OUTPUT_TENSOR_NAME])
    print ("Found in optimized")

Answer 1

尝试将input_node_names和output_node_names更改为字符串列表。为我解决了一个类似的问题。

还可能冻结或剥离未使用节点的图表，从而摆脱输出，这看起来像一个标识节点。