ECSE-626B: Statistical Computer Vision


Time: Mondays and Wednesdays 10:00-11:35am
Room: ENGTR 2100
Instructor: Tal Arbel
Office: MC-425
Office Hrs: Mondays 2:00-3:00pm or by appointment
Phone: 398-8204

Grader: Pierre-Luc Bacon



Course description:

Computer vision has always been concerned with solving ill-posed problems. For example, given an image of a scene, we are asked to determine the underlying structures that generated it. The problem is difficult in that information is lost when the three dimensional world is projected onto a two dimensional image. As a result, different scenes could give rise to the same image. To illustrate the point, consider the example below. On the left, you can see a toothpaste tube placed on a table. On the right is one particular 2D image of it acquired from a camera at a particularly unfortunate viewpoint - facing the top of the tube. Note that several different objects could give rise to that same image. Identifying the toothpaste tube from that image is difficult even for humans.

toothpastetoothpaste-2D

  




In this course, we explore the conjecture that the world is uncertain and therefore should be described through the language of probabilities. We will describe several standard problems in computer vision and explore probabilistic inference methods for their solution. The goal is to provide the student with the necessary tools to be able to apply these techniques to other computer vision research problems and to many other areas.


Background requirements

This course is designed for graduate students pursuing interests in the areas of computer vision, robot vision and artificial intelligence (e.g. machine learning, decision-making). It is not meant as an introductory course in computer vision and, as such, does not provide a broad overview of the field. That being said, we will describe and address a wide range of problems in computer vision, but the particular focus will be on solutions based on probability and information theory.

The course assumes some mathematical background in probability and statistics, linear algebra, and calculus. Students should be familiar with basic techniques in image processing and optimization. Finally, students should be comfortable programming in Matlab.


Topics to be covered: (subject to change)

The course will describe a wide variety of difficult and open problems in computer vision. These include (but are not limited to): shape-from-shading, stereo vision, object recognition, face classification, active vision, image registration and alignment, and image segmentation. We will explore probabilistic solutions to these problems by studying the following:

·  Standard regularization and Bayesian regularization approaches to solving ill-posed vision problems

·  Sequential Bayesian methods

·  Bayesian inference

·  Model-fitting - parametric and non-parametric

·  Active data selection

·  Information theory

·  Pattern recognition

·  Markov Random Field (MRF) models

·  Principal component analysis and eigenmodels


Reading material

Readings will include journal and conference papers in computer vision as well as sections from the following textbook:

·  David J.C. MacKay, "Information Theory, Inference, and Learning Algorithms". 

Some suggested background reading:

·  David A. Forsyth, Jean Ponce, Computer Vision: a Modern Approach, Prentice Hall, 2002.

·  T.M. Cover and J.A. Thomas, Elements of Information Theory, Wiley & Sons, New York, 1991.

·  Stan Z. Li, Markov Random Field Modeling in Image Analysis , Springer-Verlag, 2001.


Grading:

·  Assignments: 20%

·  Midterm literature review: 10%

·  Final project:

·        Document: 30%

·        Presentation: 20%

·  Final test: 20%


Useful computer vision links:

·  The Computer Vision Home Page

·  Annotated Computer Vision Bibliography

·  CVonline Compendium of Computer Vision


Academic Integrity:

"McGill University values academic integrity. Therefore, all students must understand the meaning and consequences of cheating, plagiarism and other academic offences under the Code of Student Conduct and Disciplinary Procedures (see www.mcgill.ca/students/srr/honest/ for more information).”

"L'université McGill attache une haute importance à l’honnêteté académique. Il incombe par conséquent à tous les étudiants de comprendre ce que l'on entend par tricherie, plagiat et autres infractions académiques, ainsi que les conséquences que peuvent avoir de telles actions, selon le Code de conduite de l'étudiant et des procédures disciplinaires (pour de plus amples renseignements, veuillez consulter le site www.mcgill.ca/students/srr/honest/)."

 

Right to Submit in English or French Written Work that is to be Graded:

 

“In accord with McGill University’s Charter of Students’ Rights, students in this course have the right to submit in English or in French any written work that is to be graded.”

 

"Conformément à  la Charte des droits de l’étudiant de l’Université McGill, chaque étudiant a le droit de soumettre en français ou en anglais tout travail écrit devant être noté (sauf dans le cas des cours dont l’un des objets est la maîtrise d’une langue)."

 

Students with Disabilities:

 

"If you have a disability please contact the instructor to arrange a time to discuss your situation. It would be helpful if you contact the Office for Students with Disabilities at 514-398-6009 before you do this."

 

Course Evaluations:

 

"End-of-course evaluations are one of the ways that McGill works towards maintaining and improving the quality of courses and the student’s learning experience. You will be notified by e-mail when the evaluations are available on Mercury, the online course evaluation system. Please note that a minimum number of responses must be received for results to be available to students."

 

Please Note:

 

"In the event of extraordinary circumstances beyond the University’s control, the content and/or evaluation scheme in this course is subject to change."