Interactive Monte Carlo Denoising
using Affinity of Neural Features

result viewer

Single-frame denoising

We collect results from our algorithm as well as previous image-space denoising methods at various number of samples per-pixel (spp) on single-frame denoising task.

To facilitate the comparison, each link below will lead you to an interactive page in which you can flip between several scenes, and methods. The lower the spp, the noisier the input.

The interactive viewer also allows you to zoom (using the mousewheel) and move within the image (by dragging).

4spp 8spp 16spp

Video denoising

Below you can find the video denoising results for 4spp inputs.

In the video, FPS indicates only the performance of the respective denoiser. That is, time it takes to render 4spp input is not included in the FPS calculations.

ONND is from the OptiX 7.2 SDK, and unfortunately the official implementation does not account for temporal stability.

video (4spp)