Calling TensorFlow Python Program from C++

This Demo will show how to call an pre-trained imagenet model to predict picture in C++.

├─ prediction.cpp                 % C++ file
├─ vgg_model.py                   % TensorFlow vgg model   
├─ makefile                       % Compile file
├─ little_demo                    % An simple Demo
└─ test_pic/                      % Test pictures
	├─ cat.jpeg     
	├─ puzzle.jpeg
	└─ tiger.jpeg

The python script

Firstly, we need an python script which let us to load VGG model & use it to do predict op.

You can test the script vgg_model.py by using the following cmd:

(deeplearning) bg@bg-cgi:~/Desktop/C_python$ python3 vgg_model.py 
Using TensorFlow backend.

...
...
...

 Please input picture file to predict: test_pic/cat.jpeg
 Predicted: [('n02124075', 'Egyptian_cat', 0.93183666)]

It will download weight file vgg19_weights_tf_dim_ordering_tf_kernels.h5 and imagenet_class_index.json to ~/.keras/models the first time.

Compile

use Makefile

Using make all to Compile your cpp file.
Make sure your python's version is correct.

bg@bg-cgi:~/.keras/models$ locate Python.h
/usr/include/python2.7/Python.h
/usr/include/python3.5m/Python.h

so I use the following cmd to compile

g++ -std=c++11 -Wall -O3 prediction.cpp -lpython3.5m -L/usr/include/python3.5m/ -I/usr/include/python3.5m/ -o pred_test

use Cmake

I just write a CMakeLists.txt, which can help you to compile your cpp code.
So just use the following cmd:

mkdir build && cd build
cmake ../
make
cd ../

Then you will see a executable file classifier, just use ./classifier to run the program.

Usage

Makefile

make all to compile, then ./pred_test to run c++ program.

Cmake

use ./classifier to run the program after building.

Then you can see it works:

 ===========> START CALL PYTHON SCRIPT <===========
 ===========> 1st CALL <===========
 Please input picture file to predict: huhu
 file not exist!
 ===========> 2nd CALL <===========
 Please input picture file to predict: test_pic/cat.jpeg
 Predicted:  [('n02124075', 'Egyptian_cat', 0.93183666)]
 ===========> 3rd CALL <===========
 Please input picture file to predict: test_pic/tiger.jpeg
 Predicted:  [('n02129604', 'tiger', 0.82598984)]
 ===========> 4th CALL <===========
 Please input picture file to predict: test_pic/puzzle.jpeg
 Predicted:  [('n03598930', 'jigsaw_puzzle', 0.99813461)]
 ===========> CALLING FINISHED <===========
(deeplearning) bg@bg-cgi:~/Desktop/C_python$

Feel free to contact me if you have any question.

BIGBALLON/CPP-Call-Tensorflow