real adaptive subdivision

[bestman8]

adaptive subdivision more like cycles has. i think it will help a lot with displacement.

---

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

[juangea]

LuxCore already has adaptive subdiv, what do you miss about it?

[bestman8]

i mean more like cycles adaptive subdivision so it is not the hole mesh but the parts closest to the camera

[Theverat]

You will have to describe in more detail how this feature should work, and illustrate it with images. Or supply an example .blend file showcasing the Cycles subdivision.

[bestman8]

this is just a plane with adaptive subdivisions on it here is a image of how cycles does it luxcore subdivides the whole object but with cycles it depends on where the camera/viewer is and will subdivide the objects according to that

[bestman8]

a few probably better images

[bestman8]

here is that file so you can see it your self.zip go in camera view maybe move the plane so it updates and then move out camera view to see it from above like in the last images.

[juangea]

Very interesting, do you have the same example but with the LuxCore result?

[bestman8]

i dont because i dont know how to visualize it with lux but there should the entire mesh be even

[Dade916]

Actually, it is already supported:

It is just adaptive simplification (i.e. you first uniformly subdivide and than adaptive simplify): https://forums.luxcorerender.org/viewtopic.php?f=5&t=1523

[juangea]

ah! it was the arlequin node what I was looking for to see the result :)

Dade is that the same that happens with adaptive subdivision?

I ask this becauseif subdivision is not adaptive you have to generate billions of polygons to simplify them afterwards, not sure it that's what is expected by adaptive subdivision or of a progressive subdivision would be more efficient avoiding the need to generate super big amount of geometry before.

Maybe everything happens under the hood and while we generate it in two steps the engine is just avoiding to generate the geometry that will be useless :)

[Theverat]

In my opinion, Subdiv + Simplify is a very poor substitute for real adaptive subdivision. The difference in memory requirement and compute time is immense. So I would also welcome "true adaptive subdivision" for single meshes in LuxCore.

Also, it might make sense to combine adaptive subdivision with displacement into one step, so the adaptiveness can better match the requirements of the displacement. For example, it could subdivide further in areas where the displacement texture has fine high-frequency detail, and subdivide not at all in areas where the displacement texture is uniform.

[bestman8]

so should i rename it to real adaptive subdivision?

[Dade916]

The subdivision is done by Pixer's OpenSubdiv.

[juangea]

Mmmh then may be easy to implement true adaptive subdivision since it’s part of the Pixar OpenSubDiv library :)

https://graphics.pixar.com/opensubdiv/docs/subdivision_surfaces.html

[Dade916]

@juangea, the links is for "Curvature-adaptive tessellation of B-spline surface" (see the label): it is for tessellation of parametric surfaces, it is quite different topic from the one asked by Bestman8.

[juangea]

Yeah, you are right, I just looked into it with more detail :)

[mambobanda]

View Dependent Adaptive Subdivision / Tessellation would be awesome. Especially if it takes the distance of the camera into account. Found one paper about it, there are several different ones out there. This is just for example.

http://hhoppe.com/proj/vdrpm/

And if you want to get even more fancy, there's subdivision that takes features into account.

"Feature-adaptive GPU rendering of Catmull-Clark subdivision surfaces" (Theres a paper out there).