Tuesday, November 9, 1999
4:00 PM (reception following)
Room 35-225
LIDS Colloquium
Abstract
Given a number of cluttered pictures containing cars (or faces, or shoes) can a model of `car' be learnt without human supervision? I will present a solution to this problem. It is based on learning probabilistic object models composed of `parts' and `shape'. Learning happens in two phases: at first a large number of promising candidate features are selected; then a shape density and the model features are estimated simultaneously. I will present a number of experiments demonstrating excellent detection and generalization performance. ------- Joint work with Markus Weber, Max Welling and Wolfgang Einhauser
|
Modified: Nov 8, 1999|
Current events|
Your comments
and inquiries are welcome.