First CVPR held in Washington, DC

A field that needed a shared venue

Computer vision is the branch of computing that tries to make machines derive information from images and video: locating objects, recovering the geometry of a scene, following movement from frame to frame. Its methods borrow from image processing, statistics and machine learning, and progress depends on researchers being able to test their methods against the same problems and compare results. A regular conference gives a field that meeting point.

The first meeting

The Conference on Computer Vision and Pattern Recognition (CVPR) was first held in 1983 in Washington, DC ^[1]. The first edition was organised by Takeo Kanade and Dana H. Ballard ^[1].

The name joins two traditions that the organisers brought into one room. Computer vision is the problem of interpreting what is in an image; pattern recognition is the statistical machinery for deciding which class a piece of data belongs to. In practice the two are hard to separate — a system that finds an object in a photograph still has to classify it — and a single conference made the connection explicit rather than leaving vision work and classification work in separate communities.

Sponsorship and open access

From 1985 to 2010 the conference was sponsored by the IEEE Computer Society ^[1]. Since 2012 it has been co-sponsored by the IEEE Computer Society and the Computer Vision Foundation, which provides open access to the conference papers ^[1].

The later arrangement changed who could read the work. Conference papers are normally locked behind subscription proceedings; with the Computer Vision Foundation providing open access, the papers presented at the conference can be read by students, engineers in industry and researchers in places without a library subscription. For a field that moves quickly, that shortens the path from a result being announced to other people building on it.

How selective it is

CVPR generally has acceptance rates below 30% for all papers, and below 5% for oral presentations ^[1]. The gap between the two numbers is the visible part of the review process: a paper can be judged sound enough to appear in the proceedings and still not be chosen for a spoken slot in the main programme. Authors read the oral acceptance rate as a signal of which results the programme committee considered most significant, and the overall rate as a signal of how crowded the submission pool is.

Position in the field

CVPR 2016 was the 29th edition of the conference ^[2]. That count is a reminder that the meeting did not run without interruption from its first year, and that its history is bound up with the wider growth of the field: as vision shifted from hand-designed features towards learned models, the conference became the place where those shifts were argued out, paper by paper.

What a reader should take from the 1983 meeting is modest but real. A research area that had been spread across workshops and journal articles got a named, recurring occasion with two organisers willing to run it, a city to host it and, within a couple of years, an institutional sponsor behind it ^[1]. Everything else — the acceptance rate, the open-access proceedings, the place in a young researcher's calendar — was built on that.