Invited Talks

Philip H. S. Torr
Oxford Brookes University

Biography: Philip Torr did his PhD (DPhil) at the Robotics Research Group of the University of Oxford. He left Oxford to work for six years as a research scientist for Microsoft Research, first in Redmond USA in the Vision Technology Group, then in Cambridge UK founding the vision side of the Machine learning and perception group. He is now a Professor in in Computer Vision and Machine Learning at Oxford Brookes University.

Philip Torr won several awards including the Marr prize (the highest honour in vision) in 1998. He is a Royal Society Wolfson Research Merit Award Holder. Recently he together with member of his group have won several other awards including most recently an honorary mention at the NIPS 2007 conference for the paper P. Kumar, V. Kolmorgorov, and P.H.S. Torr, An Analysis of Convex Relaxations for MAP Estimation, In NIPS 21, Neural Information Processing Conference, 2007, Best Paper at Conference for O. Woodford, P.H.S. Torr, I. Reid, and A.W. Fitzgibbon, Global Stereo Reconstruction under Second Order Smoothness Priors, In Proceedings IEEE Conference of Computer Vision and Pattern Recognition, 2008. Recent SIGGRAPH on VideoTrace work with the University of Adelaide has been featured extensively in the press and led to a spin out PunchCard.

He was involved in the algorithm design for boujou released by 2D3. Boujou has won a clutch of industry awards, including Computer Graphics World Innovation Award, IABM Peter Wayne Award, and CATS Award for Innovation, and a technical EMMY. He continues to work closely with this Oxford based company as well as other companies such as Sony and Sharp.

Towards Global Energy Models for Scene Understanding

Scene understanding is one of the fundamental aims of computer vision. A lot of work has been in to individual elements of the problem, such as designing the best sliding window detector to recognize a particular object, or on 3D structure recovery. Less effort has been put in to combining these elements to produce an understanding of the whole scene.

In this talk I outline work over the past few years at the Oxford Brookes computer vision group on developing new Hierarchical Conditional Random Field graphical models of the scene understanding problem. These allow a formulation of the problem as one of energy minimization, with a probabilistic interpretation, and we show that for certain classes of models this can be solved very efficiently.

This class of models features very large clique sizes (sometimes tens of thousands) and yet remains computationally tractable, and is a natural fit to the scene understanding problem allowing for a natural blending of top down and bottom up information.