Welcome to ICCV 2013, the first ICCV to be held south of the Tropic of Capricorn! ICCV attendance has steadily increased over the past decade, reflecting the growing importance of computer vision both as an academic research field and a source of ideas ripe for entrepreneurship. As the conferences grow, the logistics of organizing them increases also, and we want to thank many people for their hard work in making ICCV 2013 a success.
The Program Chairs – Kyros, Philip, Steve and Yi – heroically handled a very large number of submissions, driving their area chairs, in turn, to manage many hundreds of reviewers to submit thousands of thoughtful and penetrating reviews. They dealt with the occasional problems that arose professionally and sensitively; the result is an excellent program that reflects the exceptional quality of work being done in our field.
The conference, as usual, includes a number of excellent tutorials and workshops, and we want to thank the tutorial chairs (Derek, Rob and Marshall) and the workshop chairs (Pushmeet, Deva and Kyoung Mu) for working with the tutorial and workshop proposers to craft a balanced and relevant program.
Computer vision conferences enjoy a healthy level of financial support from companies that have a strategic interest in our field. ICCV has received support from many of these companies (NICTA, ANU, Kitware, NREC, Raytheon, Siemens, Adobe, Google, Megvii, Disney, Facebook, OMRON, Sarnoff/SRI, NVidia), and we are very grateful to them for their contributions that allow us to support important activities and some of the conference prizes. We are especially grateful to Anthony Hoogs and Anthony Dick, who personally contacted and negotiated with these companies on behalf of the conference.
Sydney is a large (and expensive) city, and low cost accommodations, especially for students, are critical for allowing many people to attend the conference. Simon Lucey and Massimo Piccardi were the conference local arrangement chairs, and they did a superb job of identifying accommodations convenient to the conference venue at low cost to participants.
ICCV is hosting a large number of demos that highlight computer vision technologies that have been transitioned to industry or are poised to do so. We want to thank Antonio Robles-Kelly who single-handedly organized the exhibit and demo component of the conference.
The conference web page was developed by Ryan Farrel and Grant Van Horn and we thank them for their hard work. Jian Zhang was the publications chair, and he successfully chased down authors and was quick to answer all questions about deadlines, formats, etc.
We also want to thank Gerard Medioni who served as finance chair, but more importantly brought a wealth of experience and good humor to the table and saved us from ourselves more than once.
Finally, we want specially to acknowledge the IEEE Computer Society and the Computer Vision Foundation (CVF) who have supported the conference in many ways. Student travel scholarships were funded by the IEEE PAMI Technical Committee who also underwrote the conference. For the first time, ICCV also benefits from the sponsorship of CVF. All papers are indexed by IEEE, but in a first for ICCV, all papers will also be hosted for open access by CVF.
Nobody gets paid for any of this! So, we want to close by thanking all those members of the community who have committed many person-weeks of their time to the success of ICCV 2013.
Larry Davis and Richard Hartley, General Co-Chairs
Welcome to Sydney, Australia and the 14th IEEE International Conference on Computer Vision (ICCV). In addition to the main four day program of oral and poster presentations, ICCV 2013 has 25 co-located workshops, 10 tutorials, and demos and exhibits.
We received 1629 submissions to the conference (a record for ICCV). Of these, 41 (2.5%) were accepted as orals, and 413 (25.4%) as posters, for an overall acceptance rate of 27.9% See http://www.iccv2013.org/decisions.php for a more detailed breakdown of acceptance rates as a function of research area and review scores.
To handle the review process, we invited 50 leading vision researchers to act as Areas Chairs (ACs) and recruited 1152 expert reviewers from the computer vision community, with a maximum of 11 papers per reviewer (most had far fewer). Our candidate reviewer list consisted of the CVPR 2013 reviewers, with the addition of area chairs from the last several computer vision conferences and editorial board members from leading vision journals. We were extremely impressed with the diligence of the reviewers across the board. Out of nearly 5000 reviews (each paper had three reviews), only a couple dozen were not returned---an extraordinarily small number. For these papers, we called on a set of senior “emergency reviewers,” not in the regular reviewer pool, who graciously agreed in advance to handle one or two papers at the last minute with a day or two turn-around time.
We used the CMT conference management service sponsored by Microsoft Research to manage the submission and selection of papers from beginning to end. The AC meeting was faciliated by a set of new scripts to give detailed and individualized per-AC and per-panel paper/review summaries distributed via dropbox.
After the submission deadline, we asked ACs to bid on preferred papers to review, and combined the resulting bids with the automated Toronto Paper Matching System (TPMS) to assign papers to ACs. TPMS, develped by Charlin et al. [UAI 2011], suggests matches between papers and reviewers (ACs, in our case) based on bag-of-words descriptors extracted from the PDF files of submitted manuscripts and representative publications by each potential reviewer. The ACs in turn used the results of a TPMS matching of papers to reviewers to help them determine the potential reviewers for each of their assigned papers, from which the CMT system automatically selected three non-conflicted reviewers per paper.
Conflicts were handled differently this year. In the past, if you had a collaborator at (e.g.,) MIT, you had to declare a conflict with the entire institution, thereby ruling out all reviewers at MIT. Furthermore, the need to self-declare conflicts led to omissions, as many authors failed to list all their collaborator’s institutions. Instead, this year, we instructed authors to only list their own institutions (not collaborators’ institutions), and we auto-detected collaborators using an online publication database (DBLP). For each author A, we marked a reviewer R as a conflict if A and R co-authored a paper in the last 3 years, and one of them was first author. In addition, reviewers could notify us in case they noticed (or suspected) conflicts. However, we received only a handful of these, indicating that the auto-detection strategy worked well.
Reviewers were given four weeks to complete their reviews, at which time the ACs stepped back in to vet the reviews for quality (initiating discussions with reviewers, where necessary) before they were released to the authors. After the author rebuttals were collected, a second discussion phase ensued.
New this year, we asked reviewers to rate papers both before AND after the author’s rebuttal. Furthermore, in the latter (post-rebuttal) rating, we removed the borderline option, and instead required reviewers to commit to recommending the paper as reject, (weak) poster, or (weak) oral. While this required more work on the part or reviewers, the vast majority completed both ratings, and the ACs found them extremely useful in decision making.
The AC meeting was held at Oxford, to make decisions on all but the clear rejects. Clear reject decisions were made by the individual AC handling each paper, and verified by a second AC. Roughly two thirds of submissions made it to the meeting. On the first day of the AC meeting, every paper was discussed by an AC triple, with the goal of making accept/reject decisions. Each AC was part of two different triples, one that met in the morning, and the other in the afternoon, to provide more fine-grained handling of conflicts. Only papers not rejected at the end of day one made it to day two. Day two focused on poster/oral decisions, and all oral candidates were discussed by a panel of about a dozen ACs. Each AC was part of two different panels, one that met in the morning, and the other in the afternoon (again, for more fine-grained conflict handling). The Program Chairs served as the panel chairs. By the end of the meeting, the ACs were asked to produce detailed consolidation reports to justify all their decisions.
The Program Chairs and General Chairs did not submit any papers to ICCV 2013, allowing them to work without any direct conflicts throughout the review process. Additionally, the ACs were excluded from any decisions associated with papers that they authored or from their affiliated institutions or close collaborators. Triples had no conflicts, and AC panels had papers authored by the ACs in that panel. Soft (institutional or collaborator) conflicts were allowed in panels, with all conflicted ACs leaving the room when each paper was discussed. The double-blind nature of the CVPR review process was thus strictly maintained throughout the review process.
ICCV 2013 is co-sponsored by IEEE and The Computer Vision Foundation (CVF). The proceedings of ICCV 2013 are being published in USB drive form. All papers in the main conference and associated workshops will be indexed by the IEEE, and available through the IEEE Computer Society Digital Library and under IEEE Xplore.
We wish to thank the other members of the Organizing Committee, the Area Chairs, Reviewers, Authors, and the CMT team for the immense amount of hard work and professionalism that has gone into making ICCV 2013. Our thanks also go to the organizers of CVPR 2013 for their helpful advice and support. We are grateful to the sponsors as well. Finally, we wish all the attendees a stimulating, informative, and enjoyable conference.
Kyros Kutulakos, Yi Ma, Steve Seitz, and Phil Torr, Program Co-Chairs