Computer Vision Summit CVPR has just (virtually) taken place, and like other CV-focused conferences, there are quite a few interesting articles. In fact, more than I could possibly write individually, so I have gathered the most promising from the big companies here.
Facebook, Google, Amazon and Microsoft all the articles shared at the conference – and others as well, I’m sure – but I’m sticking to the big hitters for this column. (If you are interested in the articles deemed most deserving by participants and judges, the nominees and the awards are listed here.)
Redmond has the most interesting papers this year, in my opinion, because they cover several real needs that are not obvious.
One documents this shoebox that we or perhaps our parents filled with old 3x5s and other film photos. Of course, there are already services that help with this, but if the photos are crumpled, torn, or otherwise damaged, you usually get a high resolution analysis of that damage. Microsoft has created a system to automatically repair these photos and the results look very good.
The problem is as much to identify the types of degradation from which a photo suffers as to correct them. The solution is simple, write the authors: “We are offering a new triplet domain translation network by exploiting real photos as well as massive synthetic image pairs.” Amazing, no one has tried it before!