Polygon edges in VrayDisplacement

[Joelaff]

Anyone know how to get rid of the polygon edges you see in the displacement here

I have tried subdivison displacement, which sometimes helps, but doesn't make it go away.
Lowering the edge length helps, but it is still visible in motion.
Keep Continuity has no effect
Texmap min 0 or -100 had no effect (Cellular is 0-1.0 AFAIK)
Light Shadow bias up or down (to zero even) has no effect
Secondary Ray Bias has no effect
Using Object space for the Cellular texture, or a UV Map makes no difference.
The object is from Alembic. However, I re-UVMapped it with a cylindrical map and it had no effect
Turbosmooth/MeshSmooth/OpenSubdiv did not help
Applying a Smooth modifier with autosmooth did not help
Welding nearby vertices with an Editpoly did not help
Polygons are not doubled. There is no inside Geo.

The object shades perfectly smoothly without displacement.

Max 2023 Windows 10 VRay 6.00.21 build 31863

Thoughts? Thanks.

[Joelaff]

Here is the simplified scene file.

[Joelaff]

Note VRayPluginTex vs. Max Cellular did not help.

[Joelaff]

I *think* Using Subdivision Displacement mode and enabling Smooth UVS with Preserver Map Bounds set to All seems to have done it... But why would this have any effect if I was using a Cellular map set to Object XYZ?

[fixeighted]

I was always under the impression that long polys were to be avoided ideally, so maybe it's as simple as that.
I retopologised and that cured it in this simple example.

[Joelaff]

I appreciate you looking at it. I don't think it's the polygon shape specifically (those aren't very long, honestly). I tried connecting some edges to make them more square and still had the issue. What retoplogy method did you use?

I am not saying there isn't something wrong with the model, because I think there is. It came from Maya via Alembic and something is fishy. I tried resetting and collapsing and still got this result.

[Joelaff]

The retopology modifier and a fresh UV map does seem to fix it, as you say. But what is the problem with it? Is it really just the non-square polygons?

I’ve used displacement on much worse geometry and rarely encounter this issue. I am pretty sure I have seen it before though.

[fixeighted]

I couldn't say techinically why it sometimes happens on seemingly arbitrary meshes, though I do remember that
years ago, when I didn't know how displacement worked optimally, I'd try to use it for bricks e.g. The geometry was full of tris/quads of very different sizes...very bad, and I always
got this stuff happening.
So I think it does make a difference to have generally at least even sized faces, even though they may be a bit messed up.

An algorithm thing maybe?
Maybe it looks at, in your example, the long polys and judges them to be fine around the circumference of the mesh, as the distance is equal, though creates a sort of
falloff between them over the vertical length that creates this issue.....but that is purely a guess

[Joelaff]

I am hoping Chaos will chime in with a technical description of what is happening

[Joelaff]

Wondering if it is procedural textures being sampled poorly or something. I will try to use an image instead.

Any thoughts from the Chaos crew?

[hermit.crab]

It does seem to be reliant on the "squareness" of the topology since it works fine after a Retopology modifier. Since 3D Mapping is only tessellating, whereas Subdivision uses a subdivision scheme to edges (another way of subdivision), it can bypass the "retopology" step. This process is internal and optimized for VRayDisaplcement. Otherwise, to get even more technical a developer is required.

[Joelaff]

I would be very interested in any more details. Obviously we want displacement to work with as many geomtries as possible, not just perfect clean quads. I wonder if it is only the procedurals. Still need to test that. Like maybe the 3d sampling of the procedural is (for speed?) based on the original vertices (possibly lerped) rather than the displaced ones and that is causing issues.

[kosso_olli]

We ran into the same issue today. Using retopo fixed it, but it would be better if this could be handled straight out of the box.