LuxCoreRender Forums

oh..lets try, thanks

Hi Dade, sorry i was wrong, we used 0.05 as you suggest.
Prefiltering was activated too.
We use bidir only.

Can you post a simple test scene ? You use BiDir so it is a different situation from 99.99% of Blender renderings.

I can post only a luxcore scene, not a blender scene. Is that fine?

daros wrote: Fri Jul 02, 2021 4:27 pm I can post only a luxcore scene, not a blender scene. Is that fine?

Yes, it is.

Here it is:
https://drive.google.com/file/d/1LCHemc ... sp=sharing

daros wrote: Mon Jul 05, 2021 8:44 am Here it is:
https://drive.google.com/file/d/1LCHemc ... sp=sharing

The scene is to complex to be hand edited but there is something quite wrong in the Albedo AOV:


The Albedo is mostly black and it is going to cause a lot of problems to OIDN. For instance, why the walls albedo is black ? It may be something in the way you define the materials but there too many objects in the .scn to isolate only the walls: post a simpler scene with just the walls and a light source and I can check what is going on.

There are also some quite strange thing in the .scn:

- there is a pitch black infinite light, in an indoor scene;

- all the "splitfloat3", gamma correct, "makefloat3" thing is really terrible for performance. It is done at run time, it is very expansive and really bad for performance. You should use color spaces: viewtopic.php?f=5&t=3446
And usually, you want also to reverse gamma correct the inputs (and gamma correct the outputs).

here is simplified scene:
https://files.fm/u/her8zm3z3

bartek_zgo wrote: Mon Jul 05, 2021 1:39 pm here is simplified scene:
https://files.fm/u/her8zm3z3

The source of the problem was in Disney material glossiness: Albedo AOV was handling Disney as a specular material (i.e. like a mirror) because the roughness was 0.0.

Disney material can be matte, glossy, specular and everything in between so it is hard to classify. I added an heuristic so Disney is considered specular if and only if:

1) metallic or specular (average) value is >= 0.5
2) roughness < 0.05 (i.e. glossiness threshold)

Note: I'm now also using the square root of Disney roughness so the value is directly comparable with Glossy2, Metal, etc. roughness (this breaks the compatibility with the past for everything related glossiness threshold).

The end result is:


If you compare the new Albedo AOV with the one in my previous post: it has a lot more details and it should help Intel OIDN a lot more.

Thanks you a lot dade ,the improvement is strong.

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




In this scene i wanted to test the transparency improvements in oidn however the result suggest me we did something wrong.

previous oidn latest oidn attached you can find the scene