Baking Bevel in Cycles

[AndreasResch]

I've created a series of comparison renderings between Cycles and Luxcore both using Normal Maps and Bump Maps. I've used some rock pebble maps from CC0 with lots of details. There are versions with just the sphere and the HDRI in the background and there are versions with a ground plane below the sphere. Especially the versions with the ground plane show the issue with the black areas rather well. As mentioned before, there's also something going on with the bump maps. They look quite different and there are issues with dark areas and hard edges as well. Maybe this helps to find the culprit of the issue.

Here's a layered Photoshop file which is great to compare the renderings ... http://www.argfx.at/upload/NormalBump_Comparison_01.zip

And here are the side-by-side JPG's for viewing the images in the forum.

NORMAL MAP


NORMAL MAP WITH GROUND PLANE


BUMP MAP


BUMP MAP WITH GROUND PLANE

[Dade]

Actually, the lighting of Cycles looks totally wrong to me, the bright border of the sphere edges doesn't make sense, it is pulling light out of nowhere

[kintuX]

2. edit: indeed, after further testing it looks like there's discrepancy in one of the axes(?)

Thanks for confirming this. Maybe this could be investigated. If you need any assets, let me know.

Cheers.

Well, I ran out of time to investigate it further.
But very nice to see it solved.

About the latest, "darkening" so to speak, I guess it's the lack of Microfacet Based Normal Mapping feature in LuxCore.
e.g.


ps
another great reference is appleseed's "Google Summer of Code 2020 Report - Microfacet Based Normal Mapping" by Lovro Bosnar

[AndreasResch]

Actually, the lighting of Cycles looks totally wrong to me, the bright border of the sphere edges doesn't make sense, it is pulling light out of nowhere

It's just a single environment HDRI. Here's the comparison without the normal/bump maps enabled.

[Dade]

BTW, very first rendering of a bevel cube with v2.6 code:

Notice as it is true geometry (it is done with "capsules": https://www.shadertoy.com/view/Xt3SzX), not fake shading normal tricks (not even triangles). It also doesn't require to trace additional rays.
Based on paper: "Efficient Rendering of Rounded Corners and Edges for Convex Objects" (Simon Courtin, Sébastien Horna, Mickaël Ribardière Pierre Poulin and Daniel Meneveaux) https://simoncourtin.github.io/publicat ... vex-object

[TAO]

BTW, very first rendering of a bevel cube with v2.6 code:


bevel.jpg


Notice as it is true geometry (it is done with "capsules": https://www.shadertoy.com/view/Xt3SzX), not fake shading normal tricks. It also doesn't require to trace additional rays.
Based on paper: "Efficient Rendering of Rounded Corners and Edges for Convex Objects" (Simon Courtin, Sébastien Horna, Mickaël Ribardière Pierre Poulin and Daniel Meneveaux) https://simoncourtin.github.io/publicat ... vex-object

Amazing as i can see many users ask for it. nice to see bevel features in LuxCore.

[AndreasResch]

BTW, very first rendering of a bevel cube with v2.6 code:

Damn it. We ask for a Toyota and get a Lamborghini. I hope this works well on subtle bevels as well. If it does, I will skip 2.5 and head for 2.6 Alpha straight away.

Will the normal/bump issue be investigated nonetheless?

[Sharlybg]

Fantastic

I hope this works well on subtle bevels as well

There is no reason why it shouldn't (in the paper the method seem to be just fine )

[AndreasResch]

Yeah. I've checked the papers and the video after I have written it. Looks really impressive. Can't wait to test it.

[kintuX]

Capsule for edges is just f***ing ingenious

That's why I love you Dade, always delivering amazing surprises!

PS
BTW, have you considered how to tackle and represent concave edges then?