World has never seen more photorealistic graphics than this, and it could be the future for graphics in gaming.
Enhancing Photorealism Enhancement
One of the more exciting facets of Grand Theft Auto V is how closely the game’s San Andreas resembles real-life Los Angeles and Southern California. However, a recent machine learning initiative named “Enhancing Photorealism Enhancement” from Intel Labs can drive it much further into photorealism.
The processes produced by the researchers Stephan R. Richter, Hassan Abu Alhaija and Vladlen Kolten yield a startling result: a visual look which has unmistakable resemblances to the sort of images which can be taken incidentally from the smudged front of your vehicle.
You need to see everything in action to fully understand it, but the mix of slightly washed out lights, cleaner floors and reflective vehicles simply sell the impression that you look out from a real dashboard on the real highway, even though everything’s simulated.
The Intel scientists started with made photographs and ran them via an image enhancement network. That’s pretty common for some form of image enhancement, even upscaling. The AI network also snoops the graphics card g-buffers, collecting information from the game engine about the materials, forms, and lighting in the current scene. The data is then fed into a g-buffer encoder network, which generates function tensors.
The network scores each scene’s realism using a perceptual discriminator, with actual images providing additional detail on the topic. To ensure that the network knows how to handle trees, data from the scene is collected and numbered. Data from the scene is extracted and labeled to ensure the network understands treating trees differently than cars, for example.
The team describes the process in great depth in the video, but it does take some technical knowledge. Anyone who is familiar with these words is welcome to read the team’s research article. It’s a fascinating improvement for players, as the improved graphics would undoubtedly make games more fun and practical.