hanulbogo/meanfield-matlab

MATLAB wrapper for Efficient Inference in Fully Connected CRF

A MATLAB wrapper for solving DenseCRF problems [1,2]. The code uses the c++ library provided with [2].

Getting started

• Setup a C++ compiler in MATLAB using mex -setup, you will need to have a supported compiler installed, see http://se.mathworks.com/support/compilers/index.html for a list
• To solve a general problem see examples/example.m
• To perform segmentation on the MSRC-21 database using the unary potentials from http://graphics.stanford.edu/projects/densecrf/unary/ see examples/example_MSRC.m

Included solvers

• Mean field approximation, using approximate filtering [2]
• Mean field approximation, performing all summations explicitly (slow)
• TRWS-S [3]
• Graph cuts for 2 label problems [4]

References

1. Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials.
   Conference on Neural Information Processing Systems (NIPS), 2011.
   Philipp Krähenbühl and Vladlen Koltun.

2. Parameter Learning and Convergent Inference for Dense Random Fields.
   International Conference on Machine Learning (ICML), 2013.
   Philipp Krähenbühl and Vladlen Koltun.

3. Convergent Tree-reweighted Message Passing for Energy Minimization.
   IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2006.
   Vladimir Kolmogorov.

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