One of the impressive aspects of GTA V is that the game’s San Andreas is close to the real-life of Los Angeles and Southern California. Still, a new machine learning project from Intel Labs, “Enhanced Photo Fidelity Enhancement,” can be exciting. Fidelity to practice. Realistic direction.
The game process created by researchers Stephan R. Richter, Hassan Abu Alhaija, and Vladlen Kolten produced surprising results: the visual appearance is similar to the kind of photos you might casually take from the front window of a dirty car. You have to see it to appreciate its movement. Still, the combination of slightly muted lights, softer sidewalks, and incredibly reflective cars goes to show the fact that you’re looking at the actual street from the real dash, even if everything is virtual.
Intel researchers suggest that some of the realism comes from the data set they provide for neural networks. The group provided a more in-depth and comprehensive explanation of the basic working principle of image enhancement in their document (PDF). As far as I know, the cityscapes dataset used, constructed primarily from photos of streets German, is full of details. Its brightness and brightness are different, but it can almost capture the smoothest and most interactive Google Maps street view scrolling effect I imagined. It doesn’t behave exactly like the real thing, but it looks like it’s built from natural things.
The researchers say their enhancements also exceed what other realistic conversion processes can provide by integrating geometric information from GTA V itself. The researchers say that these “G buffers” can include data such as the distance between game objects and the camera and the quality of the textures (like the shine of a car).
Although you may not see an official “realistic update” released for GTA V tomorrow, you may have played a game or watched a video that benefits from another type of machine learning artificial intelligence update. The process of using intelligent machine learning tools to break down graphics to a higher resolution appears nowhere. Still, it has been reflected in Nvidia’s Shield TV and multiple different projects that focus on updating graphics for older games. In these cases, the neural network will make predictions to fill in the missing detail pixels in lower resolution games, movies, or TV shows to achieve these high resolutions.
Fidelity may not be the only graphical goal for video games (except for craftsmanship, which looks a bit creepy), but the Intel Lab project shows that there can be as much room for growth in software as there is in software. Original GPU capabilities for new game consoles and gaming PCs.