LuxCoreRender Forums

This one is with prefiltering off...
no big difference regarding the transparencies

daros wrote: Wed Jul 07, 2021 11:27 am latest oidn
View 01 1200x600B_1.jpg

unfortunately even the areas where oidn gets lost are very visible.

The only solution I see, for this kind of problems, is to increase the samples per pixel and/or increase the rendering resolution and than scale down the result. It is easy to test, if you render at 64/128/256 samples per pixel and increase the resolution 600/1200/2400 pixels, the problem fades away.

This is a problem somewhat specific of BiDir and having to disable pixel filtering because of OIDN (Path tracing uses a different kind of pixel filtering).

I'm playing with the idea of having a very special pre-processing of Albedo and Normal AOVs for BiDir: they can be computed very very fast by a dedicated rendering step instead of having them as side product of BiDir rendering.
The problem is BiDir has a very very low samples/sec speed and this affect the quality of Albedo and Normal AOVs. If I render them in a dedicated step with a lot more samples it should improve the quality of the denoise output with a very little cost.

daros wrote: Thu Jul 08, 2021 7:10 am This one is with prefiltering off...
no big difference regarding the transparencies

View 01 600x300newprefiltering off.png

I think I can fix this problem:


See the normal AOV ? It uses the normal of the glass instead of the normal of the surfaces seen trough the glass (so it is not aligned with the Albedo AOV).
I need to have for normal AOV the same setting I have for Albedo AOV.

great!

Dade wrote: Thu Jul 08, 2021 11:08 am I'm playing with the idea of having a very special pre-processing of Albedo and Normal AOVs for BiDir: they can be computed very very fast by a dedicated rendering step instead of having them as side product of BiDir rendering.
The problem is BiDir has a very very low samples/sec speed and this affect the quality of Albedo and Normal AOVs. If I render them in a dedicated step with a lot more samples it should improve the quality of the denoise output with a very little cost.

This sounds great! Please, enhance BiDir with this special pre-processing. Many of BiDir addicts will very appreciate this enhancement.
I' am sure about this one.

Now the Normal AOV is consistent with Albedo AOV settings (compare with my previous post):


I have also added the pre-filtering of normal pass because it can now noise as albedo.

I will now check if I can get something out of the idea of a pre-rendering of Albedo and Normal AOVs.

Dade wrote: Sat Jul 10, 2021 11:20 am Now the Normal AOV is consistent with Albedo AOV settings (compare with my previous post):

n1.jpg

I'm sure I'm not reading this correctly, but is this part of the normals correct? Looks a bit strange to me.

Martini wrote: Sat Jul 10, 2021 12:28 pm I'm sure I'm not reading this correctly, but is this part of the normals correct? Looks a bit strange to me.

Yes, it is the reflection of the side (glass both reflect and transmit and Albedo and Normal AOVs show that).

Dade wrote: Sat Jul 10, 2021 12:59 pm

Martini wrote: Sat Jul 10, 2021 12:28 pm I'm sure I'm not reading this correctly, but is this part of the normals correct? Looks a bit strange to me.

Yes, it is the reflection of the side (glass both reflect and transmit and Albedo and Normal AOVs show that).

Ah of course. Thanks for clarifying

Today I have updated our luxcore to latest commit. Unfortunately I do not see any improvement with denoiser.
With high quality it looks the same. The same problems with carpet.
Old (updated on 7'th Jully): New (updated today): With low quality glass looks even worse.
Old New