>>13129591I watched the vid. the results are impressive, but comparing it to methods that only use a single image as input and calling it "state of the art" is kinda unfair. this new method they are presenting passes the ingame render, as well as a depth map, an ID map and a bunch of other things that I didn't really understand
Still, it's pretty cool and I could see some use for ingame stuff, but it's not gonna be useful for taking already existing recorded gameplay and processing it (unless you also saved all the other layers of image that this method requires)
I wonder if that's what the future of photo realistic graphics in games will be like, low poly count vertex stuff for rough geometry and visual ML shit trained with huge models stacked on top in order to make it look real
If that's the way it's going, I'm actually pretty stoked
