UTCS Colloquium-Cordelia Shmid/INRIA: "Learning Visual Human Actions from Movies" TAY 3.128, Monday, April 27, 2009

Contact Name: 
Jenna Whitney
Date: 
Apr 27, 2009 11:00am - 12:00pm

There is a signup schedule for this event (UT EID required).

Type of
Talk:  UTCS Colloquium

Speaker/Affiliation:  Cordelia Sh

mid/INRIA

Date/Time:  Monday, April 27, 2009 11:00 a.m.

Location:  TAY 3.128

Host:  Kristen Grauman

Talk T

itle:  "Learning Visual Human Actions from Movies"

Talk Abstract:

We address the problem of recognizing natural human a

ctions in diverse and realistic video settings. This challenging but import

ant subject has mostly been ignored in the past due to several problems one
of which is the lack of realistic and annotated video datasets. Our first

contribution is to address this limitation and to investigate the use of mo

vie scripts for automatic annotation of human actions in videos. We evaluat

e alternative methods for action retrieval from scripts and show benefits o

f a text-based classifier. Using the retrieved action samples for visual le

arning, we next turn to the problem of action classification in video. We

present a new method for video classification that builds upon and extends

several recent ideas including local space-time features, space-time pyram

ids and multi-channel non-linear SVMs. Furthermore, we show how to mine re

levant context information and use it to improve action recognition. We fin

ally apply the method to learning and classifying challenging action classe

s in movies and show promising results.

This is joint work with

I. Laptev, M. Marszalek and B. Rozenfeld.