UTCS Colloquium-Dhruv Batra/Carnegie Mellon University: "Graph-Structured Discrete Labelling Problems in Computer Vision," TAY 3.128, Monday, May 10, 2010, 10:00 a.m.

Contact Name: 
Jenna Whitney
Date: 
May 10, 2010 10:30am - 11:30am

There is a sign-up schedule for this event that can be found
at http://www.cs.utexas.edu/department/webeven

t/utcs/events/cgi/list_events.cgi

Type of Talk: UTCS Colloquium

Speaker/Affiliation: Dhruv Batra/Carnegie Mellon University

Date/Time: Monday, May 10, 2010, 10:00 a.m.

Location: TAY 3.128 <

/p>

Host: Inderjit Dhillon & Pradeep Ravikumar

Talk Title:Grap

h-Structured Discrete Labelling Problems in Computer Vision

Talk Abst

ract:

A number of problems in computer vision (e.g., image
segm

entation, gender classification of faces, etc) can be
formulated as

graph-structured discrete labelling problems,
where the goal is to pr

edict labels (e.g.
foreground/background, male/female) for a set of v

ariables
(e.g. pixels, faces in an image, etc)  that have some
known
underlying structure (e.g., neighbouring pixels in an image
often have related labels). This task of inferring optimal
labels o

f structured variables is typically posed as the
minimization of a dis

crete energy function over a graph, and
is NP-hard for general graphs

.

In the first part of this talk, I will describe a new
ap

proximate
inference algorithm called Outer-Planar Decomposition (OPD).

OPD
decomposes the given intractable energy-minimization problem

over a graph into tractable subproblems over outerplanar
subgrap

hs and then employs message passing over these
subgraphs to get an app

roximate global solution for the
original graph. OPD outperforms curre

nt state-of-art
inference methods on hard synthetic problems and is
competitive on real computer-vision applications.

In the seco

nd part of this talk, I will demonstrate our work
in applying this st

ructured prediction paradigm to computer
vision applications like mult

i-class segmentation, gender
classification, interactive co-segmenta

tion of groups of
related images and interactive 3D reconstruction of

objects
and scenes.

Speaker Bio:

Dhruv Batra is a fin

al-year Ph.D. student in the ECE
department at
Carnegie Mellon Un

iversity,  supervised by Tsuhan Chen. For
the past 1.5 years,

he has been a visiting student at Cornell
University. He received a Ma

sters degree from CMU in 2006,
during which he worked with Martial He

bert from the Robotics
Institute. Before joining CMU, he earned a B.T

ech from the
Institute of Technology, Benaras Hindu University.

His research interests are computer vision and machine
learning

;
specifically, learning and inference in Markov Random Fields.

He is also interested in applications of combinatorial
optimization al

gorithms to learning and vision problems.