The new ML rendering method is described as enhancing the in-game graphics by leveraging a convolutional network that has been trained on the 'Cityscapes dataset'. The results are clearly different and appealing in a real-world way. However, it is explained that the 'Cityscapes dataset' isn't derived from driving around Southern California, but mostly from driving around cities in Germany with an automotive grade camera. One unintended side effect of using the German data is that the landscapes in San Andreas become greener, and the roads smoother and less pot-holed. Other effects like the shading and lighting of cars and buildings translate very well, but I think SoCal-centric data would make the whole photorealism effort a bit sunnier.More details at Hexus.
Intel shows AI-based photorealistic rendering in GTA V
Posted on Friday, May 14 2021 @ 12:51 CEST by Thomas De Maesschalck
Intel researchers cooked up a new machine learning algorithm to provide a more realistic, photoreal-like experience in video games. Below you can see a video that demonstrates the technique in GTA V. It shows a split-screen of the training dataset and the AI-based rendering output in the video game. Most of the footage that was used to train the algorithm was captured in Germany using an auotomotive grade camera -- so it looks greener and less sunny than the original video game.