Every VR headset wastes most of its rendering effort. Your eye only resolves fine detail in a tiny central region; everything else is peripheral vision that can’t tell the difference. Foveated rendering exploits that by drawing sharply where you’re looking and coarsely elsewhere, and it’s a big part of why Steam Frame’s eye-tracked streaming works as well as it does.
A SIGGRAPH 2026 paper by Henry Kam, Colin Groth, Jenna Kang, Pratham Saraf, Qi Sun and Kenneth Chen proposes a second axis to exploit: not just where you look, but which eye sees what.
The effect
Your brain fuses two images into one percept, and it doesn’t average them. Prior work established two useful quirks: carefully designed differences between the eyes can improve perceived contrast, and when the two eyes receive images of different sharpness, blur suppression means the fused result is generally dominated by the sharper one.
The paper names and studies dichoptic foveation: blur one eye while applying a sharpening filter to the other, at the same place in the visual field. If blur suppression holds, the viewer perceives something close to the sharp version while you only paid to render detail once.
What they measured
This is a perception study before it’s a rendering technique. Participants viewed natural image stimuli in a VR headset, with interocular frequency differences at varying retinal eccentricities (how far from the center of gaze). Responses were scaled to a just objectionable difference (JOD) scale — a perceptual measure of how noticeable a degradation is, rather than a pixel-difference metric — and the authors fit a 4D model to the results.
Their conclusion is that these interocular frequency differences are well described by a fairly simple computational model, which is the part that makes it useful. A model this compact can be evaluated per-frame in a real renderer.
They then applied it to a realistic case: free exploration of 360° video in a headset, improving on a baseline foveation method.
Why it matters beyond rendering budgets
The immediate application is efficiency. Every saved millisecond of render time on a standalone headset is battery, thermal headroom and frame rate, and today’s foveation techniques are conservative because aggressive settings produce visible artifacts.
The broader point is more interesting for artists working in stereo: the two eyes are separate channels, and the brain’s rules for combining them are exploitable. Most stereoscopic work treats the eyes as a matched pair that must agree, because mismatches usually cause discomfort. This is a rigorous account of one case where mismatch is not only tolerated but useful — the brain resolves it in a predictable direction.
That’s a tool, and not only for saving GPU cycles. Deliberate interocular difference is an underexplored medium in stereoscopic art, and work like this is what makes it usable rather than a guess.