University of Toronto - Spring 2008
Department of Computer Science
2523S: Object Modeling and Recognition
Lecture Time: Tuesday, 1-3pm
Foundations of Computer Vision (or permission)
This course will explore issues in computer vision from the standpoint
three-dimensional object modeling and recognition. A variety
modeling schemes will be discussed and contrasted, including object-centered
vs. viewer-centered models, physical vs. geometrical models, deformable
rigid models, and geometrical vs. functional models. In addition,
a number of
recognition algorithms will also be presented, including constrained
alignment, geometric hashing, and appearance-based recognition. The
these models must be evaluated in the context of particular object
tasks. A set of criteria will be identified with which the best
object model can
be chosen for a given recognition task. Using these criteria, including
robustness, indexing power, search complexity, reliance on pose determination,
model coverage, etc, a number of the leading recognition results in
will be evaluated.
Evaluation: This year, the course will be run in a seminar-like format,
two students will present two research papers (one each) per week. The
course grade will be
based primarily on a class project (80%), which can
take the form of a
software system or a comprehensive survey paper. The
remainder (20%) of the grade will come from class participation, including
the student paper presentations. Students may choose
their own project topic
or be assigned one. Project proposals are due before class on February 1.
Students will subsequently report
their progress in a set of biweekly
progress reports and will hand in a
final report at the last class.
There will be no required textbook for the course; copies of published
papers will made available online or provided to the students.
Click here for a schedule of papers we'll be covering, and click here to see
the current student presentation schedule. Keep in mind that depending
on the direction taken by the discussions, I may adapt the outline.