Temporal AI Denoising

Guys, is there a temporal AI denoiser ? Cause i have the feeling that AI is not temporally used properly at a. For ex. Neat Video or Resolve have amazing temporal results compared to Intel OIDN and i think AI could improve on that. Are there any papers on this ?

3 Likes

The original Blender denoise does have a sort-of temporal mode that I believe was used for Spring, but it’s a post-process thing that uses temporally adjacent frames as additional inputs in the denoising process. It has to be invoked from the Python console explicitly to process a series of frames with the appropriate data channels included. It does not take into account motion vectors or anything fancy like that, so it seems of limited utility to me, but the Spring guys apparently needed it. Here’s a REDDIT discussion about it.

I don’t know what the state of support for this is in recent Blender versions post 2.80, but I will guess that it may still work. Whether this does any better job than the purely non-temporal OIDN or Optix high-tech AI denoisers I don’t know.

There’s a great news here:
https://developer.blender.org/D11442

2 Likes

I’m rendering an exr sequence overnight to test, just wondering about this part: “Requires building with the OptiX 7.3 SDK to enable.” - is this still necessary or is there a build that will allow me to do this? I assume it takes the output directory and adds a denoised channel into the existing exrs?

Tried this this morning but getting the error:

“Error: Could not find a render layer containing denoising info”

I rendered out a multilayer .exr with the denoising data checked off in the view layer passes menu. What am I doing wrong?

EDIT: Seem like I had Open Image Denoiser selected and not Optix…

EDIT EDIT: Just tried a simple scene and had Optix as the denoiser, same result.