Inference in Markov Random Fields
In this work we are investigating decomposition methods that do not decompose the graph structure of the initial problem but e.g. use subproblems that relate to labels. One advantage of these methods is the potential ability to take into account some types of high-order cliques and some global constraints on the solution.
A. Osokin, D. Vetrov, and V. Kolmogorov. Submodular Decomposition Framework for Inference in Associative Markov Networks with Global Constraints, In Computer Vision and Pattern Recognition (CVPR), June 2011. pdf
D. Vetrov, A. Osokin. Graph Preserving Label Decomposition in Discrete MRFs with Selfish Potentials. Proceedings of International Workshop on Discrete Optimization in Machine learning (DISCML NIPS), December 2011. pdf
In this work we focused on an energy that penalized the number of different labels actually used in the current labeling of MRF. We found some interesting theoretical intepretations for the energy, developed an algorithm, and tried it for geometric multi-model fitting (motion segmentation, homography estimation), image segmentation and compression.
Have you spotted a typo?
Highlight it, click Ctrl+Enter and send us a message. Thank you for your help!
To be used only for spelling or punctuation mistakes.