Believe it or not, GTA 5 has been with us for almost eight years now. However, it was after the arrival of its PC version, in 2015, when the game became a creative tool for many modders. Generally such modifications have focused on giving it a more realistic visual aspect, but you can be sure that no proposal has been as surprising as the one you will meet today.
Intel Intelligent Systems Lab, the internal team of the Californian company that experiments with artificial intelligence solutions, used GTA 5 to show the progress of its technology. The project, as they explain, basically consists of improving photorealism. In the case of GTA 5, whose graphics aspire to display a “realistic” virtual scenario, artificial intelligence aims to modify images to make them look more natural. In other words: to make it more real.
Now, we say “images” because Intel’s AI does its work in real time. That is, frame by frame. It is for this reason that some minimal visual inconsistencies can be observed. Despite the above, the final result is simply spectacular. We have never had the opportunity to see GTA 5 -or any other Grand Theft Auto- with this level of realism. You can see for yourself below:
Intel points out that its technology also relies on GTA 5’s graphics engine, as it provides essential information required by the AI. For example, the depth relative to the player’s camera, the type of surface materials and, of course, lighting data. Once this information is obtained, it is processed through a convolutional neural network to apply photorealism.
This neural network makes a comparison in record time with real-world images. In this way, it identifies what kind of photorealistic details are the most suitable to be applied to the virtual scenario of GTA 5. This is where Cityscapes, a photo bank of street and vehicle photographs from Germany, comes into play.
While the experiment is impressive, video games are unlikely to take advantage of it in the near future. The reason? It requires a computer with enormous power. Nevertheless, it gives us a good idea of how artificial intelligence continues to advance in image processing.