GLOSSARY
Mesh Generation
Mesh generation is the step that produces an explicit polygon mesh from another 3D representation — a signed distance field, a point cloud, a NeRF, or a set of multi-view images.
Definition
Standard mesh generation algorithms:
- Marching Cubes — extracts a mesh from a 3D scalar field by isosurfacing. The default for SDFs and density volumes.
- Poisson Surface Reconstruction — fits a smooth, watertight mesh through an oriented point cloud. Standard in photogrammetry pipelines.
- Ball-Pivoting — rolls a virtual ball over a point cloud to connect points into triangles. Faster than Poisson, less smooth.
- Differentiable mesh extractors — DMTet, FlexiCubes — let you train a neural pipeline end-to-end with mesh output.
Why it matters
For 3D printing, mesh generation is the final step. Whatever intermediate representation an AI pipeline uses, the printer needs an explicit triangle list. The quality, manifold-ness, and watertightness of the generated mesh determine whether the part actually prints.
Different mesh extractors trade off cleanly. Marching Cubes gives you a manifold mesh but with axis-aligned stair-stepping artifacts. Poisson smooths beautifully but may close holes you wanted open. The choice matters more than the underlying neural representation in many cases.
Common confusion
"Mesh generation" in the meshing-for-FEA sense (TetGen, GMSH) refers to building volumetric tetrahedral meshes for finite-element analysis. That is unrelated to surface mesh extraction for 3D printing or rendering, though both share the word.