Spring 2011 Archive

Prof. Trevor Darrell, trevor@eecs.berkeley.edu

Spring 2011

This course will cover computer vision techniques for object and category recognition, as well as recognition of human activity from video streams. Recognition of individual objects or activities (the coffee cup on your desk, a particular chair in your office, a video of you riding your bike) or generic categories (any cup, chair, or cycling event) is an essential capability for a variety of robotics and multimedia applications. The advent of standardized datasets and evaluation regimes has spurred considerable innovation in this arena, with performance on benchmark evaluations increasing dramatically in recent years. This course will review methods that have achieved success on such datasets, and will also consider the techniques needed for real-time interactive application on robots or mobile devices, e.g. domestic service robots or mobile phones that can retrieve information about objects in the environment based on visual observation. This class will be based exclusively on readings from the recent literature, including those appearing at the CVPR, ICCV, and NIPS conferences.

The format of the course this year will primarily be discussion based, with each class beginning with a short overview of the topic by the instructor followed by detailed student-led presentations and structured critique of selected papers. All students will be expected to actively discuss each paper each week. Class size will be limited to those who have preregistered, or to 16 students, whichever is greater, to foster an environment conducive to discussion.

Each week will focus on a different subtopic of object and activity recognition, covering three to five different papers from the recent literature. These papers will be presented jointly by two or three students, one acting as a primary presenter and the other student(s) as discussant. Each student will be expected to act as presenter once and as discussant once during the term. The presenting students will choose the papers from the list suggested for that subtopic, or they are welcome to suggest other papers.

Students are expected to be involved in a related research project during the term, and be experimenting with a technique covered during the course. (Graduate students who are not actively involved in a research project outside of the course can work on a class project specific for this course or joint with another course; undergraduates who are not actively involved in a related research project are not allowed in the course.) Students will be expected to present their research progress during the term in a ten minute presentation in the last class. Grades will be based entirely on in class presentations and participation.

This course will meet once a week, Friday 10-12noon, in the 7th floor conference room (Newton room) of Sutardja Dai Hall.

THE FIRST CLASS WILL BE JAN 28th. THE INTRODUCTION CLASS WHICH WOULD HAVE BEEN SCHEDULED JAN 21st WILL HAPPEN VIRTUALLY -- PLEASE CONTACT THE INSTRUCTOR IF YOU ARE NOT ALREADY ON THE EMAIL LIST.

Prerequisites: prior Computer Vision and Machine Learning courses, or permission of instructor. Advanced undergraduates allowed only with permission of instructor and if they are actively participating in a related research project. Students should already be familiar with or be willing to learn on their own: basic image processing in MATLAB; Optic Flow; Edge Detection; Support Vector Machines; Gaussian Mixture Models; Hidden Markov Models, etc.; students must be able to read and understand at a basic level recent conference papers in the computer vision literature.