You will have seen Grand Theft Auto V realism mods earlier than, however in all probability not as efficient as this. Kotaku reports that researchers Stephan Richter, Hassan Abu AlHaija and Vladlen Koltun have developed a method that provides an uncanny degree of photorealism to the sport utilizing machine studying. The strategy not solely adopts extra pure colours, however improves reflections, tweaks street texture smoothness and in any other case provides subtleties to GTA V‘s usually hyper-vivid look.
The visible improve required a brand new method to AI enhancement. Whereas the idea of utilizing real-world footage to information the algorithms is not new, the researchers discovered that present strategies both produced artifacts (and had been often “unstable”) or had been too gradual to be usable. There was usually a large hole between the footage used to coach the AI and the in-game scenes. The brand new methodology grabs similar-looking patches to supply higher reference factors for enhancement, comparable to similar-looking vehicles and folks, whereas preserving the body charges comparatively excessive.
You are not about to put in a mod like this any time quickly. Whereas the scientists discovered their method to be the strongest and most constant of any AI-based system they’d seen, its photorealism continues to be dictated by the supply of samples. You’d want a lot, far more footage to enhance indoor scenes and in any other case see the upgrades in each side of GTA V‘s gameplay. And there is nonetheless a big efficiency hit — the unoptimized code wants half a second of inference, and that is whereas utilizing a top-tier GeForce RTX 3090 graphics card.
The builders imagine you may tightly combine their machine studying system into sport engines to hurry it up, although. In that sense, this may occasionally symbolize the way forward for sport graphics. Simply as know-how like NVIDIA’s DLSS is making 4K gaming extra viable, you may see video games routinely lean on AI so as to add a photorealistic look that might in any other case be troublesome to realize.
This text by J. Fingas initially appeared on Engadget.