This software implements the MATLAB mex-wrapper Kovtun's algorithm to compute partially optimal labelings of multilabel pairwise Potts MRFs Anton Osokin, (firstname.lastname@gmail.com) 11.08.2013 Please cite the following papers in any resulting publication: Karteek Alahari, Pushmeet Kohli, P. H. S. Torr Dynamic Hybrid Algorithms for MAP Inference in Discrete MRFs In IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2010. I. Kovtun, "Partial optimal labeling search for a NP-Hard subclass of (max,+) problems" DAGM Symposium, pp. 402-409, 2003. PACKAGE ----------------------------- ./partialOptimalityMex.cpp - the C++ code of the wrapper ./buildPartialOptimalityMex.m - function to build the wrapper ./partialOptimalityMex.m - the description of the implemented function ./example_partialOptimalityMex.m - the simple example of usage ./example_partialOptimalityMex_Tsukuba.m - example of usage on the Tsukuba stereo MRF instance ./tsukubaInstance.mat - Tsukuba instance. Potentials were precomputed by Karteek Alahari: http://www.di.ens.fr/~alahari/papers/alahari10.pdf ./projectEnergyPotts.m - a function to project an MRF instance on a smaller one by fixing some variables ./src - C++ code by Karteek Alahari (the code is modified by Anton Osokin) visit http://www.di.ens.fr/~alahari/code/eff_mlabel_ver1.zip for the original code Note, that this package include quite a lot of things but the wrapper provides interface only to the Kovtun's method ./partialOptimalityMex.mexw64 - binary file for the mex-function compiled using MATLAB R2013a + MSVC 2012 USING THE CODE ----------------------------- 0) Install MATLAB and one of the supported compilers 1) Run buildGraphCutMex.m 2) Run example_partialOptimalityMex.m and example_partialOptimalityMex_Tsukuba.m to check if the code works The code was tested using MSVC 2012 and MATLAB 2013a