Note: All lectures are to be held in the MSRI lecture hall 2850 Telegraph Avenue, second floor. Low level vision addresses the issues of labelling and organising image pixels according to scene related properties - known as intrinsic images - such as motion, contrast, relief, texture and reflectance. Such properties are hard to capture by hand-constructed models, and so there has been a considerable movement towards specifying them "generatively": models involving cascades of random variables are constructed as explanations of images. Details of the models are filled in by learning parameters from labelled training data. New inference algorithms such belief propagation, variational inference, particle filtering, min cut and others are used to apply these models to image analysis. Already some very promising results have been obtained, for example in segmentation, in stereo vision and in analysis of texture. This workshop will be a forum for some of the latest results and thinking in this area to be presented and explored. Program Schedule Monday 21st Feb 2.30-3.30: Prof Eero Simoncelli, New York University. Image statistics, efficient coding, and visual perception 4.15-5.15: Prof David Mumford, Brown University. Is there a simple statistical model of generic natural images? Tuesday 22nd Feb 09.30-10.30: Prof Michael Black, Brown University. Image statistics and low level vision 11.00-12.00: Dr Andrew Fitzgibbon, U. Oxford. Applied natural image statistics 2.00-3.00: Prof Alyosha Efros, CMU. Data-driven vision: learning by lookup 3.30-4.30: Prof Brendan Frey, U. Toronto. Using data-based parameterizations to efficient learn hierarchical models. Wednesday 23rd Feb 09.30-10.30: Prof Bill Freeman, MIT. Learning to separate shading from paint 11.00-12.00: Prof Song Chun Zhu, UCLA. From primal sketch to 2 1/2 Sketch -- shape from shading, stereo, and motion 2.00-3.00: Prof Andrew Blake, Microsoft Research. Fusion of colour, contrast and stereo for bi-layer segmentation 3.30-4.30: Dr Michael Isard, Microsoft Research. Estimating stereo and optic flow using loopy belief propagation Thursday 24th Feb 09.30-10.30: Prof Daniel Huttenlocher, Cornell. Speeding up belief propagation for low and mid level vision 11.00-12.00: Prof Yair Weiss, Hebrew University. Linear programming belief propagation and low-level vision 1.30-2.30: Prof Alan Yuille, UCLA. Beyond BP? Approximate Inference and the DLR equations 3.00-4.00: MSRI mathematics seminar for those who wish to attend 4.30-5.30: Prof Ramin Zabih, Cornell. Graph cut energy minimization algorithms for computer vision Friday 25th Feb 09.30-10.30: Prof Julian Besag and Raphael Gottardo, U. Washington. Microarray imaging with MRFs and MCMC 11.00-12.00: Prof David Donoho, Stanford. Scaling properties of higher-order image statistics: implications for edge/object detection