CVPR Sanity Preserver - First-page Visual Gists by Research Topic

Last week was CVPR - Computer Vision and Pattern Recognition, the largest and arguably the most prestigious conference for the Computer Vision community. I was there too, and as usual, there were great talks by luminaries in the field, and exciting new research by academic and industry research labs around the world.

After every such conference, I (and I am sure most attendees), are faced with the arduous task of sifting through the huge volume of papers - there were some 2000 accepted papers at CVPR this year - and finding the ones pertaining to the research topic of their interest.

So, this year, I decided to build my own tool to do so.

It is called CVPRSanityPreserver (in homage to Andrej Karpathy's great ArxivSanityPreserver), and it builds up first-page montages of the papers corresponding to a search term.

For example, searching for the term Deep (everything is Deep Learning these days), brings up a list of 76 papers, arranged in 4-paper montages, with the first page from each paper. The first couple of 4-paper montages are shown below.

This tool allows anyone to search papers at CVPR by title, and get a visual gist of those papers. I have uploaded the code (a simple python script) online. It can easily be modified to do the same for any conference.