Denoiser

Denoisers are algorithms that, given a noisy image from a ray tracing engine, produce a clean image. This can be done through multiple means, such as blurring neighboring pixels to smooth out images.

• 
  Noisy version of a render using adaptive sampling
• 
  Denoised version of the noisy forest scene

Many ray tracing engines utilize a form of Monte Carlo Integration in order to accurate approximate a scene. As a consequence however, noise is created. While increasing the number of samples can reduce the noise and eventually give a clean image, the processing time is impractical for most use cases, such as rendering an animation. Denoisers allow for reducing the amount of sampling done when rendering, while keeping a clean result.

Many denoisers, such as Intel's OIDN and Nvidia's OptiX, utilize machine learning to more accurately blur noise in portions of an image without losing too much detail from a uniform blur.

One common issue with denoisers, namely non-temporal denoisers, is that they can cause flickering issues in animation, due to artifacts that aren't stable across multiple frames.