Post

is it a good idea? unclear! but I think separating "generate surface points" from "stitch them together into the output mesh" is an interesting kernel of an idea

My immediate question is if you can turn the back half into a robust reconstruction algorithm for solids given a point cloud sampled by some sensor that bounces off surfaces

Oh man. I published some very old work in the area 20 years ago and this area is a mine field. It’s a fractal of corner cases all fighting you. In the middle of really nasty optimization, geometry/topology methods while also dealing with numerics. It’s a nightmare

It does not seem directly relevant but since you’re a modeling nerd I thought I would pass along this neat paper I’ve been using to build my CAD kernel. It’s BREP not implicit but perhaps thought provoking. bsky.app/profile/sequ...

Ah yes, "We have simplified BREP intersection by reducing it to mesh intersection", a famously easy problem 😛 (this looks great, thanks for sharing it)

Have you seen the m-rep niche of b-rep research, e.g. www.mattkeeter.com/projects/mrep/? I have a vague feeling that there's a robust BREP kernel hiding somewhere in there, but it's hard to surmount the Wall Of Math