A team of Intel researchers has taken on the difficult task of making GTA V more realistic for AI intervention. The impressive results still call into question the utility of hyper realism in video games.
Is a more realistic game necessarily more beautiful? This is the question that was asked by Intel engineers themselves An artificial intelligence capable of converting images of GTA V into photorealistic rendering.
Due to this small technical feat made possible For a team work from Intel Intelligent Systems Lab, A cell dedicated to research on artificial intelligence, machine learning and robotics at Intel. As an introduction to his research articleThree authors state that ” Photorealism has been the goal for video game graphics for half a century “, And recent advances in the world of artificial intelligence allow for better” Bridging the gap between synthetic and real images »
Mercedes Everywhere, Realism Nowhere
To prove this, the researchers therefore modified the images of GTAstar, an extremely popular game from Rockstar, to give them a more realistic look, and to show their results in photos and videos.
To make GTA V more realistic, the Intel team fed a homemade artificial intelligence With the Cityscape database, Which has lots of pictures of urban environment. The machine had the task of dissecting the game’s images into multiple “layers”. This allows you to have a thorough and accurate view of many aspects of the rendering, such as the illumination of the scene, the back and distances of the camera, even the brightness of 3D objects. To superimpose images from Cityscape and GTA, AI has therefore produced more realistic images.
This operation may seem simple on paper, but it actually required a great deal of work from researchers, as the machine learning methods traditionally used in such an operation produced very random results here. One of these methods, for example, added Mercedes jewelry to all cars, simply because the cityscape images are mainly from Germany where many Mercedes cars roam the streets.
Should video games be hyper realistic?
To address these issues, it was decided not to analyze the full context of an image to AI (which is traditionally done in these cases, helping the machine understand its surroundings), but Small pieces to segment each picture in the game, to match them with the actual pictures. Thus, small elements of the GTA landscape, such as a cyclist, or a view of a highway dividing line, will be compared with similar photos on the cityscape.
The result obtained by Intel is impressive. The pictures produced from his experiments seem very close to our reality in fact. However, they are often less flattering to the eye. The extreme contrast of 3D rendering, the saturated colors of the weights, the swarm of details give way to a more realistic vision, but less rich.
Over the years, video games have developed a distinct grammar and universe to which we have become accustomed. Bright green trees and luminous clouds do not strike in video games as they might in life. Applying Hyperalistic treatment to GTA loses part of its gaming identity, much like “Troubled Valley” in humans.
Continues in video