├── .gitignore
├── README.md
└── keras_to_tensorflow.py


/.gitignore:
--------------------------------------------------------------------------------
1 | keras_model/*
2 | tensorflow_model/*
3 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # keras_to_tensorflow
 2 | kerasで出力したmodelファイル(.h5)をtensorflowのmodel形式(.pb)に変換するスクリプト
 3 | 
 4 | [amir-abdi](https://github.com/amir-abdi)さんが公開されている.ipyenvコードを.pyに変換してから改良。  
 5 | <https://github.com/amir-abdi/keras_to_tensorflow>
 6 | 
 7 | * keras modelをコマンドライン引数から選択
 8 | * 出力ファイル名は入力ファイルを元に定義
 9 | 
10 | 
11 | ## 使い方
12 | **kerasとtensorflowが動く環境を用意**
13 | 
14 | ```bash
15 | pip install keras
16 | pip install tensorflow
17 | ```
18 | 
19 | **以下のようなディレクトリ構造に**
20 | 
21 | ```
22 | .
23 | ├── README.md
24 | ├── keras_model
25 | │   └── keras_model.h5
26 | ├── keras_to_tensorflow.py
27 | └── tensorflow_model (存在しない場合は自動生成)
28 | ```
29 | 
30 | **実行**
31 | ```bash
32 | # sample
33 | python keras_to_tensorflow.py keras_model/keras_model.h5
34 | ```
35 | 
36 | 
37 | ## Original README
38 | ### keras_to_tensorflow
39 | General code to convert a trained keras model into an inference tensorflow model
40 | 
41 | The notebook ```keras_to_tensorflow```, is a sample code which loads a trained keras model, freezes the nodes (converts all tensorflow variables to tensorflow constants), and saves the inference graph and weights into a protobuf file (.pb). This file can then be used to deploy the trained model for inference. During freezing, other nodes of the network, which do not contribute  the tensor that would contain the output predictions, are prunned. This results in a smaller, optimized network.
42 | 
43 | The code on how to freeze and save keras models in previous versions of tensorflow is also available. Back then, the freeze_graph tool (```/tensorflow/python/tools/freeze_graph.py```) was used to convert the variables into constants. This functionality is now handled by ```graph_util.convert_variables_to_constants```
44 | 


--------------------------------------------------------------------------------
/keras_to_tensorflow.py:
--------------------------------------------------------------------------------
 1 | #!/usr/bin/env python
 2 | # coding: utf-8
 3 | 
 4 | """
 5 | __doc__
 6 | General code to convert a trained keras model into an inference tensorflow model.
 7 | """
 8 | 
 9 | __author__ = "Haruyuki Ichino"
10 | __version__ = "1.0"
11 | __date__ = "2017/08/20"
12 | 
13 | print(__doc__)
14 | 
15 | 
16 | import sys
17 | 
18 | from keras.models import load_model
19 | import tensorflow as tf
20 | import os
21 | import os.path as osp
22 | from keras import backend as K
23 | 
24 | from tensorflow.python.framework import graph_util
25 | from tensorflow.python.framework import graph_io
26 | 
27 | 
28 | # Set parameters
29 | if len(sys.argv) != 2:
30 |     print("Usage: python keras_to_tensorflow.py [keras model file path]")
31 |     sys.exit(1)
32 | keras_model_path = sys.argv[1]
33 | num_output = 1
34 | write_graph_def_ascii_flag = True
35 | prefix_output_node_names_of_final_network = 'output_node'
36 | keras_model_name = keras_model_path.split("/")[-1].split(".")[0]
37 | tensorflow_graph_name = keras_model_name + '.pb'
38 | 
39 | # 出力ディレクトリの準備
40 | output_dir = './tensorflow_model/'
41 | if not os.path.isdir(output_dir):
42 |     os.mkdir(output_dir)
43 | 
44 | 
45 | # Load keras model and rename output
46 | K.set_learning_phase(0)
47 | keras_model = load_model(keras_model_path)
48 | 
49 | pred = [None]*num_output
50 | pred_node_names = [None]*num_output
51 | for i in range(num_output):
52 |     pred_node_names[i] = prefix_output_node_names_of_final_network+str(i)
53 |     pred[i] = tf.identity(keras_model.output[i], name=pred_node_names[i])
54 | print('Output nodes names: ', pred_node_names)
55 | 
56 | 
57 | # [optional] write graph definition in ascii
58 | sess = K.get_session()
59 | if write_graph_def_ascii_flag:
60 |     f = tensorflow_graph_name + '.ascii'
61 |     tf.train.write_graph(sess.graph.as_graph_def(), output_dir, f, as_text=True)
62 |     print('Saved the graph definition: ', osp.join(output_dir, f))
63 | 
64 | 
65 | # convert variables to constants and save
66 | constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph.as_graph_def(), pred_node_names)
67 | graph_io.write_graph(constant_graph, output_dir, tensorflow_graph_name, as_text=False)
68 | 
69 | print('Saved the TensorFlow graph: ', osp.join(output_dir, tensorflow_graph_name))
70 | 


--------------------------------------------------------------------------------