file submission for denoiser learning

Wed Jun 20, 2018 4:52 pm

Hi, perhaps the title is not very clear

Is there a place where we can submit files for which denoiser fails ? not clear ??

Wed Jun 20, 2018 7:18 pm

You can drop screenshots here, and if we can get the ORBX from you we can add it to the training set. If you send us ([email protected]) an email with "DENOISER SAMPLE" and screenshots + link to the ORBX on dropbox or similar, we can consider it for the next training dataset.

Wed Jun 20, 2018 7:33 pm

ok, thanks ! I find denoiser fails more behind specular materials. On a simple pool with caustics, denoiser do Strange things.

Thu Jun 21, 2018 9:57 am

HHbomb wrote:ok, thanks ! I find denoiser fails more behind specular materials. On a simple pool with caustics, denoiser do Strange things.

If your any scene has caustics, reflection and refractions visibled..., image denoised is different to and more caustics/refractions incorrected than reference (in 3D virtual or in 3D real life). Bad denoised is NORMAL because caustics non-dispersed or dispesed, dispersions... are IRREGULAR.

AI denoiser isn't available for PMC kernel yet until new last versions XB or final v4. But we're recommended "progressive photon mapping" (PPM) with GI as new kernel for caustics but only final v4.