Abstract

A large variety of computer vision tasks can be formulated using Markov Random Fields (MRFs). Except in certain special cases, optimizing an MRF is intractable, due to a large number of variables and complex dependencies between them. In this project, we develop new algorithms to perform inference in MRFs, that are either more efficient (in terms of running time and/or memory usage) or more effective (in terms of solution quality), and make steps towards incorporating sophisticated MRF optimization algorithms in an end-to-end learning framework.

Publications

Fast and Differentiable Message Passing on Pairwise Markov Random Fields.
Zhiwei Xu, Thalaiyasingam Ajanthan, and Richard Hartley.
Asian Conference on Computer Vision (ACCV), November 2020. (oral)
[pdf] [arxiv] [talk] [code] [bib]

Generalized Range Moves.
Richard Hartley, and Thalaiyasingam Ajanthan.
arXiv:1811.09171, November 2018.
[pdf] [arxiv] [bib]

Efficient Relaxations for Dense CRFs with Sparse Higher Order Potentials.
Thomas Joy*, Alban Desmaison*, Thalaiyasingam Ajanthan*, Rudy Bunel, Mathieu Salzmann, Pushmeet Kohli, Philip H.S. Torr, and M. Pawan Kumar.
SIAM Journal on Imaging Sciences (SIIMS), November 2018.
[pdf] [arxiv] [code] [bib]

Memory Efficient Max Flow for Multi-label Submodular MRFs.
Thalaiyasingam Ajanthan, Richard Hartley, and Mathieu Salzmann.
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), February 2018.
[pdf] [arxiv] [talk] [code] [bib]

Efficient Linear Programming for Dense CRFs.
Thalaiyasingam Ajanthan, Alban Desmaison, Rudy Bunel, Mathieu Salzmann, Philip H.S. Torr, and M. Pawan Kumar.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), July 2017.
[pdf] [supp] [arxiv] [poster] [talk] [code] [bib]

Memory Efficient Max Flow for Multi-label Submodular MRFs.
Thalaiyasingam Ajanthan, Richard Hartley, and Mathieu Salzmann.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016.
[pdf] [supp] [poster] [talk] [code] [bib]

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.
[pdf] [arxiv] [poster] [code] [bib]

Thesis

Optimization of Markov Random Fields in Computer Vision.
Thalaiyasingam Ajanthan.
PhD Thesis, May 2017.
[pdf] [talk] [bib]