Efficient Solvers for Multi-label MRFs/CRFs

People

Introduction

Our software uses the following publicly available implementations:
a. Dynamic Graph Cuts, version 2
b. Tree-reweighted max-product message passing algorithm, version 1.3.

If you use this software for research purposes, you should cite the following papers in any resulting publication.

"Reduce, Reuse & Recycle: Efficiently Solving Multi-Label MRFs"
Karteek Alahari, Pushmeet Kohli, Philip H. S. Torr
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008.

"Efficiently Solving Dynamic Markov Random Fields Using Graph Cuts"
Pushmeet Kohli, Philip H. S. Torr
In International Conference on Computer Vision (ICCV), 2005.

"Convergent Tree-reweighted Message Passing for Energy Minimization"
Vladimir Kolmogorov
In IEEE Trans. Pattern Analysis and Machine Intelligence (PAMI), Oct 2006.

"An Experimental Comparison of Min-Cut/Max-Flow Algorithms for Energy
Minimization in Vision"
Yuri Boykov and Vladimir Kolmogorov
In IEEE Trans. Pattern Analysis and Machine Intelligence (PAMI), Sep 2004.

Papers

CVPR 2008 Paper

TPAMI 2010 Paper

BibTeX

@InProceedings{Alahari08,
  author    = "Alahari, K. and Kohli, P. and Torr, P.~H.~S.",
  title     = "Reduce, Reuse \& Recycle: Efficiently Solving Multi-Label {MRF}s",
  booktitle = "Proceedings of IEEE Conference on Computer Vision and Pattern Recognition",
  year      = "2008"
}

@Article{Alahari10,
  author    = "Alahari, K. and Kohli, P. and Torr, P.~H.~S.",
  title     = "Dynamic Hybrid Algorithms for MAP Inference in Discrete MRFs",
  journal   = "{IEEE} Trans. Pattern Anal. Mach. Intell.",
  year      = "2010",
  volume    = "32",
  number    = "10",
  pages     = "1846--1857"
} 

Code

The code can be downloaded from here (~2MB zip). A full readme file is available here.

Dataset

The datasets used in our TPAMI 2010 paper are available here (~54MB tgz).

Example

We explain the usage with the stereo computation problem as an example. Tsukuba stereo pair (courtesy of the University of Tsukuba) and the corresponding unary costs are available in images/. See the example here.

Acknowledgements

This work was supported by the EPSRC research grants EP/C006631/1(P) and GR/T21790/01(P), the IST Programme of the European Community, under the PASCAL2 Network of Excellence, IST-2007-216886. P. H. S. Torr was in receipt of Royal Society Wolfson Research Merit Award.

License & Disclaimer

Various copyrights apply to this release. It is understood that all persons using this release will adhere to the terms and constraints invoked by each author's copyright. The license file is available here.