Post Denoiser == NVidia AI Denoiser?
dan_thomas
Posts: 92
I am little confused about Post Denoiser in Rendering Settings (Filtering). Some people say the Post Denoiser is NVidia AI Denoiser but others telling they separately use NVidia AI Denoiser, implying Post Denoiser is not NVidia AI Denoiser. Could you please clarify whether it's true or not? Thanks.
Comments
There are scripts for running it separately also.
https://taosoft.dk/software/freeware/dnden/
Thanks for the info. But it would be nice if DAZ studio automatically do that. So do you agree Post Denoiser is not AI denoiser? If they are different, can you tell me how different?
I don't know which version of the Denoiser we get in this version of Iray. nVidia have been tweaking the Denoiser included with the Optix SDK over the years.
Whichever version it is in the current Iray, it does collect data with each iteration, and uses that data to make predictive Denoising passes using AI.
So, yes, it is the nVidia AI Denoiser.
Like @prixat says, the NVidia denoiser is updated regularly and both DAZ and Declan who is making the standalone scripts are usually updating their software accordingly, though not necessarily at the same time so the versions may differ at times.
That's all I wanted to know. Thank you all!
I prefer to use the denoiser in post work but it does not imply that I consider that the denoisers are different. I think both are basically identical, but you cannot consider that using the same denoiser after rendering or using it during the render process will lead to identical results. Rendering is an iterative process where the lighting on every surface somehow becomes a new light source. During rendering, when the denoiser triggers, the surfaces will have a denoised aspect and it will impact the illumination of the other surfaces.
But the fundamental reason why I personnally use the denoiser in post work, is that it allows to keep the denoised AND the original image. If you consider that the denoising was too aggressive (which frequently happens), you can easily create a final image by blending the original and denoised images together.
I generally render images at double resolution, then I denoise them, I reduce their size by two and I merge the resized original and denoised images together. All this post processing is done by a shell script with imagemagick and the offline denoiser in a couple of seconds per image.
This is not correct - the denoiser iss till a post-processor, it takes the current render iteration and denoises that but doesn't modify the render data so the enxt iteration is working from the un-denoised data. The point is that we can see how the denoised version looks and stop when it is where we want, rather than having to stop, save a copy, denoise it, and resume if we need more convergence.
Is there a way to beat the 10k x 5K that the settings seem to allow
Like any other property the Parameters settings allow to disable the limits. This way you can go higher, however I disabled the limits, set 20000x20000 as reseolution, but did not render it. So not completely tested. You may run into VRAM problems when going too high.
That makes sense. For now, I will use the post denoiser in DAZ. Thanks a lot for your helps, guys!