Would it be possible to create a 3D engine that renders everything in a scene with the aid of neural networking?You build the geometry of the scene and feed "types" to each face. So when rendering, the neural network uses all its training and data to render floors, walls, water, faces, etc. It would most likely give an eerie feeling to the images because of artifacts, but I don't see why this can't be done. Would also be the greatest step towards literal photorealism
>>680464https://people.csail.mit.edu/tzumao/diffrt/Nvidia want transform segmentation images to real life video.https://www.youtube.com/watch?v=ayPqjPekn7ghttps://tcwang0509.github.io/vid2vid/3D graphics spend a lot time build human skin model, light transportation, scattering and random walks technics to be massive out performance by one GAN.
Maybe, in a decade or two.