[Colloquium] TTIC Talks: Sanja Fidler, University of Toronto

[Liv Leader]

REMINDER:

When:     March 12, 2012 @ 11 a.m.

Where:    TTIC, 6045 S. Kenwood Avenue, Room #526

Who:       Sanja Fidler, University of Toronto

Title:        Learning Compositional Representations for Multi-Class Object
Detection

Abstract:

Visual recognition is a notoriusly difficult problem due to a large number
of visual classes, their high intra- and inter-class variability, as well
as complexity arising from imaging ambiguity. One of the main challenges of
artificial vision is to design scalable visual representations
with good generalization properties.

Hierarchies allow feature sharing between visual classes at multiple levels
of representation, can code exponential variability in a very compact way
and enable fast inference. This makes them potentially suitable for
learning and recognizing a higher number of object classes. In
this talk I will present a framework for learning a compositional hierarchy
for representing multiple object classes jointly and efficiently. The
approach builds object models from pixel-level up through recursive
composition and uses heuristics to learn the structure and the parameters
of the representation without supervision. Experimentally, the approach
scales sub-linearly with the number of classes, achieves good detection
performance, and is able to learn novel classes incrementally and
increasingly faster.

Host: Greg Shakhnarovich, greg at ttic.edu

-- 
Liv Leader
Human Resources Coordinator

Toyota Technological Institute Chicago
6045 S Kenwood Ave
Chicago, IL 60637
Phone- (773) 702-5033
Fax-     (773) 834-9881
Email-  lleader at ttic.edu <jam at ttic.edu>
Web-   www.ttic.edu
<http://www.ttic.edu/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20120309/aee8fff1/attachment.htm