How Foveated Animations Make Games Run Faster Without Looking Worse

What is it about?

This paper explores a simple but powerful idea: we can significantly reduce how much work a computer does in virtual crowd simulations by reducing the quality of animations without users noticing. We do this by leveraging how the human eye actually sees the world. A key fact about human vision is that we only see a small central area in sharp detail, called the fovea. As we move away from this center point, our vision becomes more general—we no longer pick up fine details, but only shapes, motion, and groups. In other words, our periphery sees "statistically" - we know what's there, but the details are lost to us. Techniques that make use of this are called foveated techniques, and the process itself is known as foveation. In short, foveation means focusing computing power where the viewer is looking, and using less detail elsewhere. This is usually combined with eye tracking hardware, usually offered by modern VR headsets, to keep track of where the users are actively looking. In this work, we apply the concept of foveation, but not to image sharpness and quality—as is usually done—but to character animations in virtual crowds. Because the eye can’t detect small motion details in the periphery, we hypothesize that characters in the viewer’s peripheral vision can have their animations updated less frequently—or even have them completely stopped—without the user noticing, thus saving precious system resources. We created crowd scenes with hundreds or thousands of characters and asked users to report what they noticed. The results were clear: most didn’t notice any difference, more so when the animations were fully stopped rather than reduced. This is because the reduced animations appeared "choppy", creating noticeable flicker in the user's periphery. In contrast, the stopped animations remained static and were therefore less distracting. In the best case using stopped animations, we managed to cut animation updates by over 99%. Alongside these performance gains, we also gained new insight into how we perceive motion in the periphery—especially how visual crowding and group movement can help hide missing animation detail.

Featured Image

Photo by CHUTTERSNAP on Unsplash

Photo by CHUTTERSNAP on Unsplash

Why is it important?

As games and overall simulations continue to evolve, hardware is slow to keep up. As such, methods to achieve higher performance and efficiency on existing hardware are always in high demand - foveation techniques always strived to achieve this by reducing the amount of unnecessary effort in the user's fovea. In other words, foveated animations, combined with acceleration techniques, should allow eye-tracking enabled devices, specifically headsets like the Vive Pro Eye, to run crowd-enabled games (for example, Real Time Strategy games) or general simulations at higher perceived qualities without the need to invest in better hardware. Our discovery into how motion is perceived in the periphery of the human eye - specifically the discovery that stopped animations are much more likely to go unobserved than reduced animations in the near periphery - also provides compelling evidence for further research in perceptual graphics, as well as our general understanding of the Human Visual System.

Perspectives