Continuous Inference in Graphical Models for Computer Vision
Mathieu Salzmann (NICTA)
COMPUTER VISION AND ROBOTICS SERIESDATE: 2013-05-16
TIME: 16:00:00 - 17:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
In this talk, I will discuss two approaches to performing inference in graphical models with continuous variables, and present some results on computer vision applications. More specifically, I will explain the idea of particle belief propagation, and show its application to non-rigid 3D shape reconstruction from images. I will also present an approach to performing inference in a graphical model whose energy is a polynomial function of the variables of the problem. I will show applications of this method to non-rigid shape reconstruction and shape-from-shading, and discuss its benefits and drawbacks with respect to particle belief propagation.
BIO:
Mathieu Salzmann received his M.Sc. degree in computer science in 2004 and his PhD degree in computer vision in 2009 from EPFL. He then joined the International Computer Science Institute and the EECS department at UC Berkeley as a postdoctoral fellow, and was later appointed Research Assistant Professor at TTI Chicago. He is currently a researcher in the computer vision group at NICTA. His research interests include different aspects of image-based scene understanding and the use optimization techniques for computer vision.
