The CMU Advanced Perception course is a graduate reading seminar, meeting once a week to discuss a set of papers covering a specific topic in computer vision. We look at historically important papers in field, as well as current papers from recent conferences and journals. By reading a mixture of both types of papers each week, we are able to trace the development of fundamental ideas that make up current-day computer vision research.


Syllabus

      


Marr Prize List

      


Class List


Course History

In the old days, the CMU Robotics Institute Advanced Perception Seminar was taught by committee, i.e. not very well. That changed when Steve Shafer took over and turned it into an intensive reading seminar that immersed students into the world of computer vision research through reading research papers published in refereed conferences and journals. A representative sample of the course syllabus is shown here

Shafer's AP Seminar Syllabus

After Steve left, Martial Hebert and Chuck Thorpe continued the course, using pretty much the same syllabus.

In Spring 1998, Bob Collins and Yanxi Liu took over the course and extensively revised the syllabus. We felt that Shafer's syllabus over-emphasized some areas (primarily those dealing with illumination and reflectance: color constancy, photometric stereo, shape from shading) at the expense of more mainstream vision topics (like edge detection or model matching). Also, newer hot topics were arising (mosaics, view synthesis) that we wanted to cover. Our syllabus for that semester is found here:

Spring 1998 AP Seminar Syllabus

Logistically, we also made critique writing a key ingredient in the course -- every student had to write a critique on one of the assigned papers, every week. This addressed our feeling that technical students at CMU don't get enough chances to practice their writing skills.

Although the class went well, we observed several problems with the Spring 1998 semester. First, class size had risen from a comfortable dozen or so in Shafer's day, to over 20. Second, reading 5-7 papers a week is a real grind, and we felt that the class had become an application in speed reading rather than thoughtful reflection. In Spring 1999, we have chosen to reduce the number of papers read each week to three. Furthermore, only two are assigned, and the third must be found by the student, thus giving them some practice at finding relevant papers. We found that having students write a critique on only one paper tempted them to only read that one paper, and at best only skim the others. This semester, student critiques have to cover all three papers (two assigned and one found) and to emphasize comparison of the strengths and weaknesses of each paper. Finally, student feedback from course evaluation forms complained that although the course was called Advanced PERCEPTION, it was only about computer vision, and ignored the other senses: auditory, tactile, olfactory. We actually did add a final class on hearing during the Spring 1998 semester, led by Kevin Lenzo from CS, and it was very well received. This semester, we are going to try to get guest lecturers for the following sensing modalities: hearing, touch sensing, and laser range imaging.


This page is maintained by the Robotics Institute at Carnegie Mellon University
Please address comments to rcollins@cs.cmu.edu