next up previous
Next: About this document

CVPR-97 Workshop Summaries



Workshop on Generic Object Recognition

The First IEEE Workshop on Generic Object Recognition was held on June 16, 1997, in conjunction with CVPR '97 in Puerto Rico. The workshop was co-chaired by Sven Dickinson (Rutgers University) and Ram Nevatia (USC). The workshop consisted of a number of invited speakers, including Ronen Basri (Weizmann Institute), Kevin Bowyer (USF), Chitra Dorai (IBM Watson), David Forsyth (UC Berkeley), Gerard Medioni (USC), Jean Ponce (UIUC), Stan Sclaroff (Boston U.), and Kaleem Siddiqi (Yale). There were some 85 paid regsitrants, making it one of the best-attended workshops at CVPR.

The workshop explored various methods for generic shape modeling and matching. Many of the speakers acknowledged the need to move beyond the CAD-based modeling schemes of the 80's and the appearance-based modeling schemes of the 90's, both more appropriate for exemplar-based recognition than generic recognition. It was argued by a number of the speakers that the application of content-based image retrieval is providing strong motivation for generic modeling and recognition, since exact models of database objects (either 3-D or 2-D) are typically not available.

The workshop closed with an excellent discussion that included many members of the audience and some lively exchanges. There will be no proceedings for the workshop, although a webpage is being prepared by the co-chairs that will include a list of papers referenced by the speakers during their presentations. Due to the success of the workshop, the co-chairs are considering the possibility of organizing a second workshop.

- Sven Dickinson and Ram Nevatia

Workshop on Content-Based Access of Image and Video Libraries

The IEEE Workshop on Content-Based Access of Image and Video Libraries was held in San Juan on June 20, following the 1997 Conference on Computer Vision and Pattern Recognition. The purpose of this workshop was to bring together researchers interested in the problems of organizing, representing, querying, retrieving, annotating, and browsing the visual content of video or image libraries.

There was a huge response to our call for papers, with 62 submissions made for the twelve presentation slots. Each submission was reviewed by at least two people, and the highest rated papers are included here. Over 100 people registered for and attended this workshop.

The papers included here reflect research in computer vision and pattern recognition to give people tools to access visual content. In particular, the domain of video and image libraries restates the typical automatic recognition problem as a semi-automatic problem, assisting a human in the system loop. The papers included here focus on various aspects of this goal, including tools for both general and specific kinds of querying, features for color, texture, shape, and scene structure, shot segmentation in video, representation with compressed data, relevance feedback, and retrieval on the World Wide Web.

This workshop also included a mid-day presentation by Bruce Croft of the University of Massachusetts and a discussion led by Rosalind Picard addressing the need for improved evaluation methods. The focus of this presentation was on things our community can learn about evaluation methods from the existing information retrieval community.

- R. Picard and F. Liu

Workshop on Nonrigid and Articulated Motion

The 1997 IEEE Workshop on Nonrigid and Articulated Motion was held last June in Puerto Rico in conjunction with CVPR'97. The workshop attempted to foster dialogue and debate through invited talks, panels and contributed papers on many aspects of nonrigid and articulated motion. A unique aspect of this workshop was the inclusion of papers on both nonrigid and articulated motion analysis and synthesis.

The topics of the accepted papers spanned the areas of computer vision techniques for human motion analysis and understanding, methods for object segmentation, computer graphics methods for modeling and animation of articulated objects, and medical image analysis techniques. The above topics, even though they do not cover all aspects of nonrigid and articulated motion, clearly demonstrate a growing interest in this area. We expect that in the years to come this interest will continue to grow and new theories and methods will be developed which will allow the efficient solution of difficult open problems in computer vision, graphics and medical image analysis.

The call for papers generated better than expected interest and we received papers from all over the world. The papers accepted for presentation at the Workshop went through a thorough reviewing process. Each paper was reviewed by at least two members of the program committee. The Program Committee consisted of 41 members from various research groups. The reviewing process was blind in order to avoid reviewing biases as much as possible. Based on the reviewers recommendations, 14 papers were accepted for inclusion in the proceedings. The workshop attendance was over 70 people, a clear sign of its success and the interest of the vision community in this area. The papers in conjunction with the two invited talks by N. Badler and J. K. Aggarwal (the workshop Chairs) and especially the panel, sparked lively discussions on many open research problems in nonrigid and articulated motion analysis and synthesis.

- D. Metaxas and I. Essa

Workshop on Undergraduate Education and Image Computation

A one-day Workshop on Undergraduate Education and Image Computation was held on the day immediately following CVPR '97. The workshop was organized by George Stockman, Louise Stark and Kevin Bowyer. The idea to have such a workshop was inspired by a panel discussion held at CVPR '96 in San Francisco. For more information on the CVPR '96 panel discussion and the CVPR '97 workshop, see the web page http://marathon.csee.usf.edu/teaching_resources.html.

An NSF grant provided support for some attendees who might normally not be able to attend CVPR or such a workshop. These attendees were mostly from smaller schools that emphasize undergraduate education, and may not have a graduate research programs. It was a very full day for workshop attendees, with twelve presentations squeezed into the day. It was also a full day for the presenters, with lots of questions and interaction with the audience.

The first section of the morning was titled "New approaches in teaching traditional computer vision courses." Two presentations addressed the use of Java tools for on-line demonstration of computer vision procedures. One of these papers was titled "Interactive textbooks," by Koryllos and Fisher from the University of Edinburgh, and the other was titled "Image processing and computer vision instruction using Java," by Moscariello, Kasturi, and Camps from Penn State University. The third talk in this session was "Teaching computer vision to computer scientists," by Bruce Maxwell from the University of North Dakota. Bruce presented a good comparison of various computer vision textbooks and how well they lend themselves to different approaches to teaching computer vision.

The second session was titled "Image computation in lower-level or core courses." George Stockman from Michigan State University spoke on "A first unit in computing with images," describing material that has been tried in a variety of settings to introduce people to computing with image data. Sudeep Sarkar and Dmitry Goldgof of the University of South Florida spoke on "Image computation in the undergraduate data structures course," describing how to teach the traditional data structures course with examples that all involve image computing. The paper "The computer vision component in a DSP laboratory," by Fernand Cohen and Athina Petropulu from Drexel, discussed lab exercises for a DSP course that are oriented toward image computation.

Ramesh Jain from the University of San Diego gave an invited presentation titled "A sequence of courses in visual computing." Ramesh's talk contained a number of interesting ideas. An idea for a specific course that stands out as worth replicating at other institutions is a course for non-majors on image computation using visual toolkit packages.

An afternoon session was titled "Image computation in upper-level / elective courses." Robin Murphy, from the Colorado School of Mines, gave a presentation on "Image computation in an upper-level elective on robotics." Milan Sonka, from the University of Iowa, gave a presentation titled "Image computation in an upper-level elective on medical imaging." And Mubarak Shah, from the University of Central Florida, gave a presentation titled "Mentoring undergrads in computer vision research projects." Each of these talks contained a number of good ideas for better teaching and learning related to image computation.

The last session of the day was titled "Experiences with NSF grant programs related to undergraduate education." Mubarak Shah was called on to speak a second time, this time emphasizing his experience with the NSF program called "Research Experiences for Undergraduates." The NSF REU program provides grants for "sites" and for "supplements." A "site" is a larger coordinated activity, but does not have to be related to other NSF support. A "supplement" is a add-on to an existing NSF award specifically for the purpose of supporting an undergraduate to be involved in the work. Fernand Cohen was also called on to speak again, emphasizing his experience with the NSF "Instrumentation for Laboratory Improvement" program. This program provides support for laboratory equipment whose primary use is in undergraduate education. This session was, as you might guess, very different from the typical conference or workshop session. Mubarak Shah and Fernand Cohen each did an excellent job of being open about their experiences and providing good advice to others getting started in the field.

Overall, the workshop was a great success. Everyone should have come away with several ideas that they can use to improve the quality of their teaching. A number of attendees suggested that the workshop should be an annual event. The time required for planning prevents the workshop from being held at CVPR '98, but plans are to have another workshop at CVPR '99. Also, Horst Bunke, the editor of the International Journal of Pattern Recognition and Artificial Intelligence, has agreed to have papers from selected presentations at the workshop appear as a special issue of IJPRAI.

-Kevin Bowyer





next up previous
Next: About this document



Sudeep Sarkar
Fri Aug 15 12:52:41 EDT 1997