DAGM 2010

Tutorial at DAGM 2010 on MAP Inference in Discrete Models

21/09/2010, 14:00 - 18:00

Many problems in Computer Vision and Graphics are formulated in form of a random filed of discrete variables. Examples in low-level vision are image segmentation and stereo reconstruction; in high-level vision object recognition, and graphics panoramic stitching and texture synthesis. The goal is typically to infer the most probable values of the random variables, known as Maximum a Posteriori (MAP) estimation. This has been widely studied in several areas of Computer Science (e.g. Computer Vision, Machine Learning, Theory), and many efficient techniques exists, which are able to obtaining accurate and reliable solutions to many problems. This has led to a significant increase in the use of random field models in computer vision.

The focus of this tutorial is twofold: a) to review and examine different types of discrete models, ranging from simple (low-connected) to complex (highly-connected); b) to review and compare different types of optimization techniques for these models.

Relationship to previous tutorials: This is a shortened version of the recent tutorials at ICCV 09 and CVPR 10.

Introduction (30 min)
Discrete Models in Computer Vision (60min)
Optimization techniques (60min)
a) Combinatorial Optimization (e.g. graph cut)
b) Message Passing (e.g. belief propagation)
c) move-making methods (e.g. alpha-expansion)
Comparison of techniques (30min)
Advanced topics: LP-relaxation, dual-decomposition (30min)

Organizers:

Carsten Rother, Computer Vision Group at Microsoft Research Cambridge, UK;
carrot@microsoft.com

Curriculum Vitae:

Carsten Rother received his Diploma degree with distinction in 1999 at the University of Karlsruhe/Germany. He did his PhD at the Royal Institute of Technology Stockholm/Sweden, supervised by Stefan Carlsson and Jan-Olof Eklundh. Since 2003 he is a researcher at Microsoft Research Cambridge/UK. He supervises several PhD students and gives frequently invited talks. His research interests are in the field of "Markov Random Fields for Computer Vision", "Discrete Optimization", and "Vision for Graphics". He has published more than 20 high impact papers (at least 10 citations) at international conferences and journals. He was awarded the DAGM Olympus price in 2009, and won the best paper honourable mention award at CVPR '05. He serves on the program committee of major conferences (e.g. SIGGRAPH, ICCV, ECCV, CVPR, NIPS), and has been area chair for BMVC ‘08, and ‘09.