Mesh Generation
Creating a 3D mesh - vertices, edges, and faces - from text, images, or point clouds.
Think of it like
Sculpting: starting with clay (text description or image), carving it into a shape with faces and edges.
Example
"A ceramic mug" -> generates a mesh with proper handle, rim, interior. Or: single photo -> mesh of that object.
How it actually works
Meshes are topologically fixed (cant easily change vertex count). Methods: predict vertex coordinates, predict which vertices connect (graph), or start with a template mesh and deform it. Text-to-3D often generates NeRF or point cloud, then extracts mesh via marching cubes. Single-image: depth + shape completion. Challenges: non-manifold geometry (mismatched topology), self-intersections. Explicit meshes are artist-friendly but hard to generate. Implicit (signed distance functions, occupancy) are easier to learn, harder to render.
For product teams
Game art, 3D printing, CAD augmentation.
For engineers
Marching cubes, Poisson surface reconstruction, or learned mesh deformation. Metric: Chamfer distance, F-score (geometric accuracy).
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome