MeshGPT

Uses a decoder-only transformer to generate triangular meshes.

CommonProductDesigntriangular meshtransformer model
MeshGPT creates triangular meshes by autoregressively sampling from a trained transformer model conditioned on geometry-aware tokens. These tokens can then be decoded into the faces of a triangular mesh. Our method generates clean, coherent, and compact meshes with sharp edges and high fidelity. MeshGPT shows significant improvements over existing mesh generation methods in shape coverage, with FID scores boosted by 30 points across various categories.
Visit

MeshGPT Visit Over Time

Monthly Visits

6376

Bounce Rate

42.21%

Page per Visit

1.5

Visit Duration

00:00:13

MeshGPT Visit Trend

MeshGPT Visit Geography

MeshGPT Traffic Sources