CS 598 - High-Level Recognition in Computer Vision

Spring 2007 by Fei-Fei Li

Course Summary

Time/Location: W 13:30 - 16:20, Computer Science Department, Room 401

Professor: Fei-Fei Li, Room 211 CS Building, (609)258-8130, feifeili [at] cs [dot] princeton [dot] edu

Graduate Coordinator: Melissa Lawson, Room 310 CS Building, 609-258-5387 mml [at] cs [dot] princeton [dot] edu

Course Requirements, Grading Policy and Prerequisites

There are three course requirements:

• do the reading and actively participate in discussion
• present one or two set(s) of papers
• course project: replicate either a paper we discuss in the class or develop a piece of original research, 8-12 pages of the standard IEEE conference format.

Grading Policy:

• Class participation (20%): discussions
• Paper presentation (30%): quality, clarity, depth, etc.
• Course project (50%): quality of the project presentation, quality of work, quality of writing, etc.

Prerequisites:

• Some knowledge of computer vision (e.g. COS 429) or graphics (e.g. COS426). But previous courses are not required.
• Good knowledge of basic probability and statistics, and linear algebra.
• Know how to program -- Matlab, or C/C++, or anything that can get you what you need for the project.
• Students with advanced knowledge of machine learning are highly welcome!

Course Description

Computer vision is concerned with the recognition and understanding of images. In recent years, our field has enjoyed some important breakthroughs in the area of high-level recognition, such as object identification, object categorization, and scene categorization. An increasing trend in the field is the utilization of advanced machine learning techniques such as graphical models, various inference algorithms, etc. This course will survey the recent developments of the high-level recognition problems by reading both classical and recent papers. We will organize the course around the important topics in high-level vision. In particular, we will put a focus on how different machine learning techniques have been applied to the specific vision questions.