News Highlight

Title: Conditional Ordinal Random Fields - Ratings vs. Classes

Speaker: Vladimir Pavlovic, (Rutgers University, http://seqam.rutgers.edu/site/)

Time: 1:30pm-3:00pm

Location: Babbio 321

Abstract:

Conditional Random Fields (CRFs) are the staple of many diets in computer vision today, from image segmentation, over image annotation, to activity recognition. The underlying assumption in those tasks is that some states, typically latent, take their values from a set of nominal categories while exhibiting some sort of local smoothness. These nominal categories typically indicate classes (e.g., foreground vs background or an object class assignment). However, in some modeling settings it is more reasonable to assume that the tags indicate ordinal categories or ratings. For instance, dynamic envelopes of sequences such as emotions or movements often exhibit intensities growing from neutral, through raising, to peak values. In this talk I will discuss a new model family, the Conditional ORDINAL Random Fields (CORFs), that explicitly models the dynamics/dependencies of ordinal categories.  I will then illustrate how these models fit a number of vision (and maybe some non-vision) settings we are accustomed to: modeling of facial emotions, activity recognition, and image annotation (or label rating, to be precise.)  Interestingly, this easily overlooked distinction between classes and ratings leads to some surprising results.  For instance, our hidden CORF for facial affect modeling achieves state-of-the-art performance on the both spontaneous and deliberate emotion recognition challenges.

Biography:

Vladimir Pavlovic is an Associate Professor in the Computer Science Department at Rutgers University. He received the PhD in electrical engineering from the University of Illinois in Urbana-Champaign in 1999. From 1999 until 2001 he was a member of research staff at the Cambridge Research Laboratory, Cambridge, MA. Before joining Rutgers in 2002, he held a research professor position in the Bioinformatics Program at Boston University. Vladimir's research interests include probabilistic system modeling, time-series analysis, statistical computer vision and bioinformatics.