One of the more impressive aspects of Grand Theft Auto V is how closely the game’s San Andreas matches the real-life Los Angeles and Southern California. However, a new machine learning project from Intel Labs called Enhancing Photorealism Enhancement can improve realism to incredible levels.
Researchers developed algorithms and machine learning of the neural network using photographs of real city streets in Germany. After that, they passed the image inputted by the game through their AI system. As a result, they got a realistic image, similar to photographs that can be taken through the windshield of a car. But in this case, we are talking about a moving image. However, you can see a combination of slightly washed-out lighting, a smoother sidewalk, and believable light reflections from cars. This creates the feeling that the user is looking at a real street with typical objects, such as houses, trees, road signs, and other cars.
Researchers at Intel Labs say their enhancement system goes beyond what other photorealistic transformation processes are capable of. It does this by integrating geometric information from GTA V itself. These “G-buffers,” as the researchers call them, can include data such as the distance between objects in the game and the camera, as well as the quality of textures, such as high-gloss cars.
A source: The verge