January 24, 2008
11:00am - 12:00pm
322 Votey Hall
Summary
Object recognition is one of the most challenging problems in computer vision. One of the main difficulties is in developing representations that can effectively capture typical variations in appearance within a large class of objects. Deformable models provide a natural approach for addressing this problem. While deformable models provide an elegant framework for representing a variety of objects, they also lead to difficult computation problems.
In the first part of the talk I will describe efficient algorithms we have developed for finding objects in images using different types of deformable models. In the second part of the talk I will consider the specific problem of detecting and localizing objects of a generic category, such as people or cars, in cluttered images. We have recently developed a new multiscale deformable part model for solving this problem. The models are trained with a discriminative procedure, using a formalism we call "latent SVM." We have used these models to implement an object detection system that is both highly efficient and accurate, processing an image in about 2 seconds and achieving recognition rates that are significantly better than previous systems.
Speaker Bio
Pedro Felzenszwalb is an Assistant Professor at the Department of Computer Science at the University of Chicago. His main research interests are in computer vision, algorithms and artificial intelligence. He received a BS degree in computer science from Cornell University in 1999. He received MS and PhD degrees from MIT in 2001 and 2003. After graduating from MIT he spent one year as a postdoctoral fellow at Cornell University.