Abstract: Object detection is one of the fundamental challenges in computer vision. In this talk I will consider the problem of detecting objects from a generic category, such as people or cars, in static images. This is a difficult problem because objects in such categories can vary greatly in appearance. For example, people wear different clothes and take a variety of poses while cars come in various shapes and colors. We have built an object detection system that addresses this challenge using mixtures of deformable part models. The system is both highly efficient and accurate, achieving state-of-the-art results on the PASCAL object detection benchmarks. We train our models from weakly labeled data using a discriminative procedure that we call latent SVM. A latent SVM leads to a non-convex training problem. However, a latent SVM is semi-convex and the training problem becomes convex once latent information is specified for the positive examples. This leads to an iterative algorithm for solving the training problem. I will also discuss a general framework for object detection using rich visual grammars. Joint work with Ross Girshick, David McAllester and Deva Ramanan
Pedro F. Felzenszwalb is an Associate Professor at the University of Chicago. He received his PhD from MIT in 2003. His main research interests are in computer vision, geometric algorithms and artificial intelligence. His work has been supported by the National Science Foundation, including a CAREER award received in 2008. He is currently serving as a program chair for the 2011 IEEE CVPR. He is an Associate Editor of the IEEE Transactions on Pattern Analysis and Machine Intelligence and an Editorial Board Member for the International Journal of Computer Vision. In 2010 he received the IEEE CVPR Longuet-Higgins Prize for fundamental contributions to computer vision and the PASCAL Visual Object Challenge "Lifetime Achievement" prize.