MatX Furnace Test

Forums Technical Discussion MatX Furnace Test

1490 12 0


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:22 a.m.

I've set up a furnace test between Karma/MatX and Octane/Universal material, with varying results. The most disappointing for me is the all-important micro-surface test with normal map, which MatX didn't do very well on. All images have a default of Diffuse = 1/Spec = 1 - with additional channels added or removed as needed.

Baseline diffuse 1/spec 1 Note: we can see Karma MatX outline here in the base test.

Edited by Danny McGrath - July 22, 2024 09:31:13

Attachments:
diffuse_spec1.jpg (360.3 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:24 a.m.

Metalness = 1

Attachments:
diffuse_spec_M1.jpg (354.1 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:24 a.m.

Metaness =1 Roughness = 1

Attachments:
diffuse_spec_M1R1.jpg (363.5 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:26 a.m.

Normal map micro-surface

Attachments:
diffuse_spec_normalmap.jpg (355.0 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:26 a.m.

Normal map Roughness = 50%

Edited by Danny McGrath - July 22, 2024 09:28:26

Attachments:
diffuse_spec_normalmap_R05.jpg (368.4 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:28 a.m.

And for an additional direct comparison, this is normal map + Roughness = 50% using the Octane Standard Surface Material

Attachments:
diffuse_spec_OR_SS_NM_R05.jpg (195.5 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:29 a.m.

Roughness = 1 Metalness = 0

Attachments:
diffuse_spec_R1M0.jpg (358.6 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:30 a.m.

And finally this is with all 3 spec layers at 1: spec/coat/sheen

Attachments:
diffuse_spec_S1C1.jpg (358.2 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 9:37 a.m.

Overall, I think MatX performed very poorly. This has been a known issue in Arnold for quite some time,due to Standard Surface Material leaking energy because of the lack of multiple scattering reflections and the use of classic normal mapping. These issues have possibly carried across to MaterialX? Even Blender at this stage has implemented multi-scatter GGX. Hopefully this will be improved in futures versions of MatX.


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 10:22 a.m.

Currently, the Rombo 3rd-party Arnold material is the only model I know of that implements microfacet normal mapping, based on this paper. I expect Octane does too, based on the results above.

https://www.researchgate.net/publication/321233843_Microfacet-based_normal_mapping_for_robust_Monte_Carlo_path_tracing [www.researchgate.net]


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 10:38 a.m.

In the meantime, I've found this in the docs, so I'll run the test again. Great news if this is what I hope it is.

Attachments:
diff.jpg (99.3 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 12:11 p.m.

It turns out that XPU doesn't support building your own materials from the BSDF nodes. They only render on CPU.

Attachments:
diffusesp.jpg (165.0 KB)


Danny McGrath: Member; 21 posts; Joined: May 2016; Offline

July 22, 2024 12:41 p.m.

The result using the Schlick BSDF - diffuse/spec/metal + normal map

This result is better than MatX Standard Surface, but still quite a way off the Octane Universal Material.

Edited by Danny McGrath - July 22, 2024 12:46:51

Attachments:
diffuse_spec_schlick.jpg (135.3 KB)

Quick Links

                    
                        Search links
                        Show recent posts
                        Show unanswered posts