Anisotropic remeshing creates flipped triangles.

[jdumas]

Hi!

I just tried to remesh the fandisk model in Graphite using remesh_smooth with 4096 target vertices and tri shape adapt set to 100. The resulting mesh seems to have a lot of "fold over" like these:

(It also produces an output that is completely flipped, but that's a different issue...).

[BrunoLevy]

Hi !
There are two things here:

• yes, for high required anisotropy, it may generate flipped triangles (this could be fixed by inserting additional vertices in the Delaunay triangulation but it is not implemented yet)
• this mesh has sharp edges, and the remesh_smooth() function works for smooth meshes only.

[jdumas]

What criteria would you use for inserting additional vertices in the triangulation for this one? Is only done after the CVT is done converging?

[BrunoLevy]

Delaunay refinement in nD space:
For each "bad" triangle, throw a ray through the circumcenter of the triangle, and compute the intersection I with the surface
If the normal to the surface at I and the normal to the triangle disagree, add I to the list of points
Recompute the restricted Delaunay triangulation (including the inserted I points)