English: A 3D model (figurine of a dancer) rendered using simple Monte Carlo integration, resampled importance sampling (RIS), ReSTIR with spatial reuse, and denoised ReSTIR. An environment map was used to simulate a scene with many lights, only using the upper half of the environment map. Only direct light was rendered (treating the environment map as a light; this light technically includes bounced light from objects visible in the environment map). All images except the denoised image are unbiased. At most 5 rays were traced per pixel in each image.

Top left: (Monte Carlo integration) 5 light samples per pixel, up to 5 rays traced to test shadowing
Top right: (RIS) 32 light samples, 1 ray traced to test shadowing; faster if light samples are cheap and tracing rays is expensive
Bottom left: Two ReSTIR spatial reuse passes (after initial light sampling using RIS), 4 additional rays per pixel
Bottom right: ReSTIR output processed using Intel Open Image Denoise

Images were rendered using C++ code and GLSL shader code using the Vukan API. Implementation used algorithms from the original ReSTIR paper. A radius of 30 pixels was used for neighborhood sampling. Some neighbors were rejected based on depth and normal vector similarity. An integer hash function was used for random number generation. Environment map sampling used an alias table. Denoising was done using the oidnDenoise.exe program from the Open Image Denoise 2.3.3 download for Windows.