This repository contains the source code of the algorithm described in a CVPR 2016 paper Pairwise Matching through Max-Weight Bipartite Belief Propagation. More details are provided on the [project page] (https://zzhang.org/featurematching.html). This packages has been tested using Matlab R2015b on CentOS 7.0 (a distrubition of Linux) x64.

If you find HungarianBP useful in your research, please consider citing:

@inproceedings{Zhang:2016:CVPR,
    author = {Zhang, Zhen and Shi, Qinfeng and McAuley, Julian and Wei, Wei and Zhang, Yanning and Hengel, Anton},
    booktitle = {CVPR},
    title = {Pairwise Matching through {Max-Weight} Bipartite Belief Propagation},
    year = {2016}, 
}

0. Prerequisites

1. [Boost] (http://www.boost.org/): Install the boost library via apt-get, yum, or compiling from scratch.

2. [Matlab] (http://www.mathworks.com/): Install Matlab.

3. Configuring HungarianBP

4. Downloading HungarianBP via

       git clone https://github.com/zzhang1987/HungarianBP
   

5. Fetching the mex code via

   	cd HungarianBP 
       git submodule init
   	git submodule update --remote	
   

6. Run compiling.m.

   	matlab
   	#inside matlab 
   		cd HungarianBP
   		compiling
   

7. Run demoCar.m and demoMotor.m to reproduce the results on the [Cars and Motorbikes Dataset] (https://sites.google.com/site/graphmatchingmethods/). Run demoCharater.m to reproduce the results on the [Chinese Character dataset] (http://www.escience.cn/system/file?fileId=62549).

Precompiled mex files for linux x64 are included. For other platforms, you can use any compilier that supports matlab and c++11 to compile the mex files.

If you have any issues (question, feedback) or find bugs in the code, please contact zhangzhen@mail.nwpu.edu.cn.