You may have seen Grand Theft Auto V realism mods before, but probably not as effective as that. Kotaku reports that researchers Stephan Richter, Hassan Abu AlHaija and Vladlen Koltun have developed a technique that adds an incredible level of photorealism to the game using machine learning. The method not only adopts more natural colors, but improves reflections, adjusts the smoothness of the road texture, and adds subtleties to GTA Vis normally a hyper-crisp look.
The visual upgrade required a new approach to improving AI. While the concept of using real-world images to guide algorithms is not new, researchers found that existing methods either produced artifacts (and were often “unstable”) or were too slow to be usable. There was often a big gap between the images used to train the AI and the in-game scenes. The new method picks up similar-looking fixes to provide better benchmarks for improvement, such as cars and cars. people of similar appearance, while maintaining relatively high frame rates.
You’re not about to install a mod like this anytime soon. Although the scientists found their approach to be the strongest and most consistent of any AI-based system they had seen, its photorealism is still dictated by the availability of samples. You would need lots and lots more footage to improve interior scenes and otherwise see upgrades in all aspects of GTA Vgameplay of. And there’s still a significant performance impact – unoptimized code needs half a second of inference, and that’s all using a GeForce RTX 3090 Graphic card.
The developers believe, however, that you could tightly integrate their machine learning system with game engines to speed it up. In that sense, it may represent the future of in-game graphics. Just like technology NVIDIA DLSS makes 4K gaming more viable, you might see games regularly relying on AI to add a photorealistic look that would otherwise be hard to achieve.
All products recommended by Engadget are selected by our editorial team, independent of our parent company. Some of our stories include affiliate links. If you buy something through any of these links, we may earn an affiliate commission.