Wednesday, June 9, 1pm, 2405 Siebel Center

"Finding and understanding multiple objects in images and videos"

Dr. Deva Ramanan, University of California-Irvine

Object recognition is one of the fundamental problems in computer vision. Classic approaches tend to focus on simplified scenarios of images containing a single object. Real-world images contain many objects defined by particular spatial interactions (people tend to ride atop bikes rather than under them). In this talk, I will focus on algorithms for finding and understanding multiple objects in images and videos. I will survey a variety of work done in our lab, from low-level tasks such as multi-object tracking, extracting the relative depth ordering of objects in the scene, to high-level tasks such as understanding actions defined by multi-object interactions. From a learning and inference point of view, such problems involve the novel application of tools such as structured prediction from machine learning and approximation algorithms from graph theory.