Towards Foveated Rendering for Gaze-Tracked Virtual Reality

摘要

Foveated rendering synthesizes images with progressively less detailrnoutside the eye fixation region, potentially unlocking significantrnspeedups for wide field-of-view displays, such as head mountedrndisplays, where target framerate and resolution is increasing fasterrnthan the performance of traditional real-time renderers.rnTo study and improve potential gains, we designed a foveated renderingrnuser study to evaluate the perceptual abilities of human peripheralrnvision when viewing today’s displays. We determined that filteringrnperipheral regions reduces contrast, inducing a sense of tunnel vision.rnWhen applying a postprocess contrast enhancement, subjectsrntolerated up to 2u0002 larger blur radius before detecting differencesrnfrom a non-foveated ground truth. After verifying these insights onrnboth desktop and head mounted displays augmented with high-speedrngaze-tracking, we designed a perceptual target image to strive forrnwhen engineering a production foveated renderer.rnGiven our perceptual target, we designed a practical foveated renderingrnsystem that reduces number of shades by up to 70% and allowsrncoarsened shading up to 30° closer to the fovea than Guenter etrnal. [2012] without introducing perceivable aliasing or blur. We filterrnboth pre- and post-shading to address aliasing from undersamplingrnin the periphery, introduce a novel multiresolution- and saccadeawarerntemporal antialising algorithm, and use contrast enhancementrnto help recover peripheral details that are resolvable by our eye butrndegraded by filtering.rnWe validate our system by performing another user study. Frequencyrnanalysis shows our system closely matches our perceptual target.rnMeasurements of temporal stability show we obtain quality similarrnto temporally filtered non-foveated renderings.