ReGIR - An advanced implementation for many-lights offline rendering
Auto-discovered
Visit resource ↗Description
ReGIR - An advanced implementation for many-lights offline rendering is a long technical post by Tom Clabault on Reservoir-based Grid Importance Resampling for next-event estimation with huge light counts.
It reviews the many-lights problem and Resampled Importance Sampling, then builds ReGIR with a spatial reservoir grid, better target functions, NEE++, BSDF resampling, spatial, temporal, and visibility reuse, bias fixes, MIS, and correlation reduction. Comparisons use scenes with on the order of 100k emissive triangles, and later sections cover grid size, envmap sampling, and determinism limits.
The audience is advanced rendering engineers who already know path tracing, PDFs, and NEE. Readers without that background will struggle with the integrals.
The post is free on GitHub Pages with a TOC, equations, and many 1-spp comparison images. No subscription is required.
It is useful when implementing or evaluating many-light sampling and wanting a full practical ReGIR walkthrough rather than a short abstract.