First you generate a proposal to the PAMI Technical Committee meeting (at least every June at CVPR or ICCV). If there are multiple proposals, there is a vote to determine the final choice. The proposal must have
- Information on the committee (at least general and program chairs, and maybe others -- tutorials, finiance, local arrangements). The program and general chairs make a lot of decisions, so choose carefully -- people who will commit to the necessary time. While being general chair and program chair may seem logical at times, it is a lot of work to do both. The General Chair may serve as the local arrangements chair. The Finiance Chair needs to understand the general cost structure. Budgets of past meetings are always available to use as a guideline.
- The location is very important. While exact hotels are not expected, at least a list of possible ones with what they say it might cost (not the final negotiated prices). For reference, maybe 1/4 of the attendees would stay in dorms, if they are convenient. Dorms increase the number of students that can (and will) come. Hotel costs are also important, many attendees are on contrained budgets that limit hotel expenses to some amount. There is vocal resistance when prices go over $125, and even some resistance at that level, and over $140 it gets noisier. For meetings in the US, cheap airfares are almost always available so that hotel costs may become the largest item.
- For location, consider resort vs. campus, air access, costs, season (while Phoenix and Palm Springs can be cheap, for CVPR in June, it is not really a good time), past meeting locations, etc.
- Many things are easier if someone local to the meeting site is taking responsibility for local arrangements. This is mostly making the conference decisions -- what A/V, from where, which room, where will posters and social events be, etc. Contractual arrangements go through the Computer Society conference people. It helps to walk the rooms to know how things will flow. It is possible to pay for this, but then you are stuck with decisions from someone not directly involved in the conference, and these services cost money.
- Associated workshops will need space the weekend before and the weekend after the meeting. Sizes of these range from 30 to 100.
Finally, well organized, compelete proposals do better than rush jobs.
The IEEE CS Conference Information Page has the official Computer Society information for what to do. This is very important. A general budget levels description is also available. It has general categories of numbers, not exact numbers to use.
Since a 1988 resolution at the PAMI-TC meeting: That regularly held meeting have a large program committee, with a small intersection with the previous year, and that a reasonable fraction of the committee meet to decide on the program.
Call for papers and Paper Deadlines
- Need plenty of lead time.
- Currently submission in "final form" is the norm. This reduces the amount of paper the chair and reviewers have to handle and forces authors to think about their paper. (Has not been followed for ICCV).
- The submission length should allow for the extra pages that we allow in the proceedings (for a fee -- a large conference income source).
- WWW pre submission -- to generate the paper number, which should be on the submitted papers (saves more time for the program chair than any other simple idea).
- WWW (FTP) submission of the actual paper. This can be useful for faster access, or for providing a quick copy to additional reviewers. Currently people are not comfortable reviewing only from softcopy. PDF is probably the best alternative for this.
- Note: It is important to consider the interaction of online pre sumbission deadlines and actual paper deadlines. At a minimum, the online submission deadline should be 1-3 days before the actual paper deadline, allowing authors to obtain the necessary paper number and print their paper with that number and then mail the paper. Also when you state a deadline of, say Tuesday, be sure to say what you mean: (Tuesday at 5:00 PM, Wednesday morning when you come in to work, midnight (GMT), or whatever is most convenient for you, the program chair.) Then set the actual paper deadline to fit with your planned schedule for sorting and distributing the papers (if you are working over the weekend, a Friday deadline for the actual paper makes sense, with a Tuesday or Wednesday (late night) online deadline. This allows for some delays in mailing, but not too many.
- Blind reviews. Three reviewers (sometimes only two have been used) The number of reviews may be an issue, more reviews tend to get more negatives and more middle results. Refer to the reviewer guidelines of CVPR 2010.
- Starting in 1994, the concept of an area chair was created. This person is responsible for reviewing all the reviews for the papers in the area to provide a consistent summary (and in 1998 selected the reviewers for the papers). There should be overlapping area chairs due to potential authorships conflicts and problems with varying numbers of papers in areas. In 1998 the area chairs suggested 8-10 others as reviewers for the area, but the assignments were not restricted to the one area.
- Committee meeting: Further review of the reviews to get a summary. An issue of whether this is done by topic or by number -- make sure the second pass is by a different person. This second pass should be done in small groups. Most of the committee should attend the meeting. Then all the papers are considered and decisions made by the whole committee. At this time the chairs sould keep track of the numbers, both oral and posters, and by topic if possible. One model is to use an initial count using the initial area chair recommendation and keep current totals.
- Chairs should have recent proceedings avaliable for aiding in reviews.
- Meeting location should allow good air connections to make it a short weekend trip. (In a major city)
- Issue of is the point to reject papers or to accept papers -- i.e. does a very strong acceptance counterbalance others that say reject?
Issues that arise
- Breadth of the committee, both by topics covered and location (where they are, or where they are from). Select people based on both -- not everyone with the same interest of everyone from the same places. If important areas are missing, or important research groups are left out, people will notice.
- Breadth of the papers represented, and uniformity of reviews by topic.
- Not wanting quotas by topic, but not excluding a topic just because the committee was "biased"
- Role of pattern recognition papers
- Application, systems papers
- Document analysis papers
- Parallel Sessions -- dislike of too many parallel sessions (i.e. the 5 parallel sessions in 1988). ICCV should be single track. CVPR at least 2, but no more than 3.
- Posters: How many, how few. (But they should always get the same proceeding space.)
- Money. Remember that workshops and conferences are separate finiancial entities. Income and expenses for these must be separate. The paperwork that keeps them separate must be maintained. This is not a problem when IEEE-CS handles the registration and pays the bills, but has been a problem with some foreign meetings where different people have handled the registration without the directions to keep them separate.
The exact schedule chages to fit the exact date, the dates of winter holidays and the preferences of those running the meeting. But some things are fixed. As an indication of options see the review workflow of CVPR 2010. See also the note above on what to use for deadlines.
The program committee must address the issue of duplicate submissions. This is hard to enforce or police. In general strive for quality, but reviewers are free to reject simply because they just read the same paper for another conference (and don't remember exactly where they read about the work) and thus feel that it is not new work. Authors should face some risk in multiple submissions.
The dates of the meetings sometimes make this impossible. Unless four months elapse between meetings, there is always an overlap problem for reviews and due dates.
From a 1988 resolution: As much as feasible, discussions on international conferences should be held at international conferences.
- People have expressed a desire for University locations (cheaper rooms for students especially, cheaper social costs, cheaper A/V, more expensive meeting rooms, sometimes more flexibility in facilities and schedules, sometimes less flexibility, easy to get any number of meeting rooms for 30-50 person workshops or tutorials).
- People have expressed a desire for resort locations. San Francisco and San Juan did very well, but MIT had the most. Atlantic City did not get good reviews. Major city hotel rates are always substantially more than small city rates. Resort hotel rates can be even worse, with fewer alternatives for cheap rooms. At a hotel, the meeting room should be free -- with enough sleeping rooms. But food is not cheap.
- It is common that resort locations win when the votes are taken at meetings on campuses, and campus locations win for votes taken at resorts.
- Reasonable hotel costs are important. Alternative hotels (or near by dorm rooms) should be considered. Alternative breakfast, lunch and dinner locations are highly desired -- i.e. not only the conference hotel. Feeding 400+ people lunch in the time allowed is difficult in a hotel.
- Internet connections are becoming more desireable. This can be difficult (expensive) outside of a university, but that is changing.
- Ground transportation should exist so that car rentals are not necessary. They may be desireable or extremely useful, but there should be options so that a car is not needed.
- Reasonable air connections are important, though not essential to be airport adjacent, it should be considered.
- The Banquet is desirable, and should be included in the basic costs so that those on strict per diems are not hurt.
- Consider politics when selecting locations: Some locations attract negative political reactions. You may support or oppose these views, but they exist and must be addressed (and you may decide to ignore them, but do that after considering the issue).
- It happed twice, but not this year: there is always the flood issue -- i.e. after selecting the site it may be washed away in a flood.