Iteratively Reweighted Graph Cut

Related Publication

This code implements the IRGC and IRGC+expansion algorithms described in the following paper

Iteratively Reweighted Graph Cut for Multi-label MRFs with Non-convex Priors.
Thalaiyasingam Ajanthan, Richard Hartley, Mathieu Salzmann and Hongdong Li.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2015.

Code Assumptions

1. The MRF energy has the following form
   $$E(x) = \sum \theta_{i}(x_i) + \sum \theta_{ij} (x_i, x_j)\ ,$$
   where $\theta_{ij} (x_i, x_j) = \gamma_{ij} \theta(|x_i - x_j|)$.
2. The code currently supports MRF with 4-connected grid structure only and
   nodes are labelled from 0 --> width * height - 1 in raw major ordering.

Contact

This code is for research purposes only, if you want to use it for commercial purpose please contact us.
Email: thalaiyasingam {dot} ajanthan {at} anu {dot} edu {dot} au

Dependencies

• Directory "alpha_exapnsion_west" contains the alpha-expansion code downloaded from http://vision.csd.uwo.ca/code/gco-v3.0.zip. We made some minor changes in the energy.h file to make it work with our IRGC+expansion algorithm. Therefore we included it with our source code.

• Directory "maxflow" contains the max-flow code downloaded from http://vision.csd.uwo.ca/code/maxflow-v3.01.zip. We made some minor changes to reduce the memory requirement of the Ishikawa graph. Therefore we included it with our source code.

Example Usage

To assist the user, example.cpp, Makefile and sample data files are provided. The following command runs IRGC+expansion algorithm on 10x10 image with 5 labels, with truncated quadratic pairwise potential - truncated at lambda = 2.

irgc.exe 10 10 5 <sample>/toy_unary_10_10_5.txt <sample>/toy_binary_4_10_10.txt TQ 2 1