The Weighted Reservoir Sampling (WRS) variant is used in ReSTIR (spatiotemporal reservoir resampling for real-time ray tracing). Which is a stochastic light tra

The Weighted Reservoir Sampling (WRS) variant is used in ReSTIR (spatiotemporal reservoir resampling for real-time ray tracing). Which is a stochastic light transport estimator with inbuilt spatiotemporal denoising. A light transport estimator is trying to figure out how much light flows through a scene (https://en.wikipedia.org/wiki/Radiance https://en.wikipedia.org/wiki/Radiance). For that it has to integrate the radiance across all the possible paths light could take, while maintaining the conservation of energy (https://en.wikipedia.org/wiki/Rendering_equation https://en.wikipedia.org/wiki/Rendering_equation). In all but the most trivial cases this integral of the rendering equation has no tractable closed form solution and solving it is thus done stochastically. The very basic idea is the Monte Carlo method (https://en.wikipedia.org/wiki/Monte_Carlo_method https://en.wikipedia.org/wiki/Monte_Carlo_method): Randomly sample as many paths as you can and average them. From there more sophisticated sampling strategies were developed over the last decades: - Importance Sampling (IS) - Multiple Importance Sampling (MIS) - Sample Importance Resampling (SIR) - Resampled Importance Sampling (RIS) - Weighted Reservoir Sampling (WRS) - And finally combining RIS and WRS into ReSTIR For a in depth read see: https://agraphicsguynotes.com/posts/understanding_the_math_behind_restir_di/ https://agraphicsguynotes.com/posts/understanding_the_math_b...