Testing visibility for every RIS candidate is expensive, especially when the candidate count grows. Deferred visibility traces a shadow ray only for the selected candidate—but that shortcut changes which information is available during resampling.

The article states the estimator conditions explicitly, then examines winner visibility, occluded reservoirs, shadow-ray cost, bias traps, and the scene-dependent variance trade-off behind the optimization.

Equal-time comparison between path tracing and ReSTIR

Read the full deferred visibility article →