|
|
ICCV 2009 Tutorial on MAP Inference in Discrete Models |
|
||
|
Speakers
New! Tutorial slides are now online! Please download either
all slides in pdf format or in ppt format . In total there are 6 parts. Note, you are welcome
to use these slides for giving talks. If you plan to do, please let the
respective author know. |
Guest Speaker
|
|||
|
Purpose of this
course Many problems in Computer Vision are formulated in form of
a random filed of discrete variables. Examples range from low-level vision
such as image segmentation, optical flow and stereo reconstruction, to
high-level vision such as object recognition. The goal is typically to infer
the most probable values of the random variables, known as Maximum a
Posteriori (MAP) estimation. This has been widely studied in several areas of
Computer Science (e.g. Computer Vision, Machine Learning, Theory), and the
resulting algorithms have greatly helped in obtaining accurate and reliable
solutions to many problems. These algorithms are extremely efficient and can
find the globally (or strong locally) optimal solutions for an important
class of models in polynomial time. Hence, they have led to a significant
increase in the use of random field models in computer vision and information
engineering in general. This tutorial is aimed at researchers who
wish to use and understand these algorithms for solving new problems in computer
vision and information engineering. No prior knowledge of probabilistic
models or discrete optimization will be assumed. The tutorial will answer the
following questions: (a)
How to formalize and solve some known vision problems
using MAP inference of a random field? (b)
What are the different genres of MAP inference
algorithms? (c)
How do they work? (d)
Which algorithm is suitable for which problem? (e)
What are the recent developments and open questions in
this field? Relationship to
tutorial at ECCV 2008 Pawan Kumar and Pushmeet Kohli had given a
half-day tutorial at ECCV 2008 on MAP estimation in Computer Vision (see link). The full-day tutorial at ICCV would cover the
following topics that were not part of the tutorial at ECCV: (a) Discussion of empirical studies of the performance of
different inference methods for various classes of random fields. (b) New material on recent advances: optimization of
higher-order functions; introducing various convex relaxation techniques; new
move-making methods. (c) More example applications to motivate the need of
different classes of functions and algorithms. Researchers and students who had attended the previous
version of our tutorial at ECCV would also find this tutorial useful and
informative. Final Syllabus 9.30-10.00 Introduction (Andrew Blake) – Part1 10.00-11.00 Discrete Models in Computer
Vision (Carsten Rother) – Part2 15min Coffee break 11.15-12.30 Message Passing: DP, TRW, LP
relaxation (Pawan Kumar) – Part3 12.30-13.00 Quadratic pseudo-boolean
optimization (Pushmeet Kohli) – Part4 1 hour Lunch break
15:00-15.30 Speed and Efficiency (Pushmeet
Kohli) – Part4 15min Coffee break 15:45-16.15 Comparison of Methods (Carsten
Rother) – Part5 16:15-16.45 Dual-decomposition (Carsten
Rother) – Part5 16-45-17.30 Recent Advances in Convex
Relaxations (Pawan Kumar) – Part6 About the Speakers |
||||
|
|
Pushmeet
Kohli
is a researcher in the Machine Learning and Perception group at Microsoft
Research Cambridge, and a post-doctoral associate of Trinity Hall, University
of Cambridge. He completed his PhD studies at Oxford Brookes University under
the supervision of Prof. Philip Torr. His PhD thesis, titled "Minimizing
Dynamic and Higher Order Energy Functions using Graph Cuts", was the
winner of the British Machine Vision Association's Sullivan Doctoral Thesis
Award, and was a runner-up for the British Computer Society's Distinguished
Dissertation Award. Before joining Microsoft Research Cambridge, Pushmeet was
a visiting researcher at Microsoft Research Bangalore. He previously worked in
the Foundation of Software Engineering Group at MSR Redmond. Pushmeet has
worked on a number of problems in Computer Vision, Machine Learning and
Discrete Optimization. His papers have appeared in SIGGRAPH, PAMI, IJCV,
ICCV, CVPR, ICML and ECCV. |
|||
|
|
Pawan
Mudigonda
is a post-doctoral researcher at Stanford University. Pawan obtained a
Bachelors of Technology degree in Computer Science and Engineering from the
International Institute of Information Technology, Hyderabad, India. He
completed his PhD studies in 2008 at Oxford Brookes University under the
supervision of Prof. Philip Torr and Prof. Andrew Zisserman. His work focuses
on combinatorial and convex optimization based solutions for problems in
Computer Vision and Machine Learning, and has appeared in several reputed
conferences and journals such as ICCV, CVPR, NIPS, ICML and IJCV. Together
with his collaborators, he won best paper awards at ICVGIP 2004, Rank
Opto-Electronics Symposium 2007, and NIPS 2007. |
|||
|
|
Carsten
Rother
received his Diploma degree with distinction in 1999 at the University of Karlsruhe/Germany.
He did his PhD at the Royal Institute of Technology Stockholm/Sweden,
supervised by Stefan Carlsson and Jan-Olof Eklundh. His thesis was nominated
for the Best Nordic Thesis Award 2003-2004, as one out of two candidates form
Sweden. Since 2003 he is a researcher at Microsoft Research Cambridge/UK. He
supervises several PhD students and gives frequently invited talks. His
research interests are in the field of “Markov Random Fields for
Computer Vision”, “Discrete Optimization”, and
“Vision for Graphics”.
He has published more than 20 high impact papers (at least 10
citations) at international conferences and journals. He won the best paper
honourable mention award at CVPR ’05. He serves on the program
committee of major conferences (e.g. SIGGRAPH, ICCV, ECCV, CVPR, NIPS), and
has been area chair for BMVC ‘08, and ‘09. He has organized
workshops on “interactive computer vision” at ICCV ’07, and
on “Color and Reflectance” at ICCV ’09. |
|||
