Rockstar Games released Grand Theft Auto V nearly eight years ago. The game was impressive at the time, but it’s starting to show its age now, especially in the graphics department. Intel’s Intelligent Systems Lab (ISL) took it upon itself to improve those graphics with a new method of making synthetic images—like those used in games—appear more lifelike.
ISL shared details about its new method in the recursively titled “Enhancing Photorealism Enhancement” paper on May 10. The described method uses “a convolutional network that leverages intermediate representations produced by conventional rendering pipelines,” as it “provides strong supervision at multiple perceptual levels” because of the way it was trained.
That training—among other things outlined in the paper—is said to offer “substantial gains in stability and realism in comparison to recent image-to-image translation methods and a variety of other baselines.” But a picture is worth a thousand words, and ISL was kind enough to provide a video showing its new (ahem) enhanced enhancement at work in GTA V:
We won’t know for certain until we test it ourselves, but the results shown in the video are stunning. Even if they were cherry-picked for the purposes of demonstrating ISL’s research, it’s still remarkable how much the group was able to improve the photorealism of a game that was released two console generations ago (and which has made the jump to successive platforms with minimal improvements along the way).
ISL shared additional data and code related to the project on GitHub. Maybe someone will be intrepid enough to use the group’s findings to improve GTA V’s graphics while everyone waits for Rockstar to finally release Grand Theft Auto VI. Modders have already taken to fixing Grand Theft Auto Online’s performance—why not make the base game look better as well?
