Brown University --
Computer Science Department Seminar
Abstract: Many problems in low-level vision can be productively approached in terms of energy minimization. Unfortunately, for many problems exact minimization of the energy is NP hard, and furthermore it is not clear whether the energy functions used are actually appropriate for the problem at hand.
In the first half of the talk (joint work with T. Meitzer and C. Yanover), I will discuss our recent success in obtaining GLOBAL optima for energy functions in stereo vision (a problem that is known to be NP hard). Our results clearly show the need for better energy functions. In the second half of the talk (joint work with A. Levin) I will discuss our attempts to learn energy functions for category-specific object segmentation. By using supervised learning, we can obtain high quality segmentations with a relatively simple model.
Brown Analysis Seminar
Department of Mathematics Colloquium
Abstract: Abstract can be seen on the Mathematics Department Seminar Page
<--- 2006 Index