The surprise to me is that one can Super Enhance JPGs and TIFFs — in addition to RAW files.
I think the RAW files give it a bit more information for it to work with, but the same principle applies to whatever sort of thing you run it on,it is used to do noise removal on phone calls and video conferences.
Essentially the machine just recognizes a pattern, says “I know how that is supposed to look at higher res” then puts that new image in place. The key to being able to make this work is the large amount of images that pass through their cloud that the machine can learn from (and some powerful GPUs in the background doing math). There are doubtless elements of this which work to identify and suppress unwanted or spurious artifacts.
I think we will see this type of computational processing find its way into cameras themselves. Maybe not for high end cameras, but if a phone cam can uprez by a factor of 4, sensors can get even smaller, or pixels can get larger and more effective in low light.
