Sign function implementation

[Q-Minh]

Hi @itsvismay ,

Was wondering how you guys handle deficient meshes as input geometry? What approaches are used for signed distance computation for inside/outside information? What about codimensional structures (thin shells, rods)? I am under the impression that they are still modeled as solids, which is reasonable, but wondering what sign functions capture those geometries.

[Q-Minh]

I see this implementation, which seems to assume watertight input, and I'll also check the code to see for myself, but wondering if you maybe have any additional info.

[itsvismay]

Hi @Q-Minh ,

On deficient geometries:

Deficient geometry works (up to the limits of the input geometry) just as well as a watertight mesh. The specific function I used to sample points in this case assumes a nice mesh; however, the method itself makes no such assumption. You can swap out the function for winding numbers (libigl has a good implementation) and use a non-water-tight mesh, and the simulation will work the same.

On codimensional structures:

Yes, they're still modeled as solids. We don't simulate true codimensionality at the moment. So far, for thin sheets, I've been using densely sampled SDFs to capture the geometry. I haven't simulated many rods, except one Bezier curve object, for which I already had sample points.

Sorry for the late reply,
Vismay

[Q-Minh]

No worries, thanks for the reply!