This is about video, but the results will surely filter down to our department(s):
Engineers are teaching upscalers to recognize different kinds of scenes, and fill in missing pixels by using neural networks. Samsung is trying to build this AI into consumer sets. TV Tech article.
Ultimately of course this depends on the quality of the AI training. But low-cost engines are getting faster, and there's a lot of training capacity available. So it's sure to grow.
Here's the thing: if this becomes the standard in a few years, and content providers start using more compr