Computer imaginative and prescient summit CVPR has simply (just about) taken place, and like different CV-focused conferences, there are fairly a couple of attention-grabbing papers. More than I may presumably write up individually, in reality, so I’ve collected probably the most promising ones from main firms right here.
Facebook, Google, Amazon and Microsoft all shared papers on the convention — and others too, I’m positive — however I’m sticking to the large hitters for this column. (If you’re within the papers deemed most meritorious by attendees and judges, the nominees and awards are listed right here.)
Redmond has probably the most attention-grabbing papers this 12 months, for my part, as a result of they cowl a number of nonobvious real-life wants.
One is documenting that shoebox we or maybe our dad and mom stuffed with previous 3x5s and different movie photographs. Of course there are providers that assist with this already, but when photographs are creased, torn, or in any other case broken, you typically simply get a high-resolution scan of that injury. Microsoft has created a system to robotically restore such photographs, and the outcomes look mighty good.
The drawback is as a lot figuring out the varieties of degradation a photograph suffers from as it’s fixing them. The resolution is straightforward, write the authors: “We suggest a novel triplet area translation community by leveraging actual photographs together with huge artificial picture pairs.” Amazing nobody tried it earlier than!