Abstract
Meshtron is a novel autoregressive mesh generation model able to generate meshes with up to 64K faces at 1024-level coordinate resolution — over an order of magnitude higher face count and 8x higher coordinate resolution than current state-of-the-art methods. Meshtron's scalability is driven by four key components: an hourglass neural architecture, truncated sequence training, sliding window inference, and a robust sampling strategy that enforces the order of mesh sequences. This results in over 50% less training memory, 2.5x faster throughput, and better consistency than existing approaches. Meshtron generates meshes of detailed, complex 3D objects at unprecedented levels of resolution and fidelity, closely resembling those created by professional artists, and opening the door to more realistic generation of detailed 3D assets for animation, gaming, and virtual environments.
Meshes Generated by Meshtron
(This could take a while to load)
More Results
Artist Meshes
AI-Generated
3D Scan
Quad Meshes
Large-Scale
Meshtron Architecture
Meshtron is an autoregressive mesh generator based on the Hourglass architecture and using sliding window attention. It exploits the periodicity and locality of a mesh sequence to achieve greatly improved efficiency.
Hourglass Architecture
The hourglass architecture assigns different compute to different mesh tokens based on their position within the sequence, as shown below:
Comparisons with Other Methods
Meshtron vs. MeshAnythingV2
Meshtron vs. Iso-Surfacing Methods
Citation
@article{hao2024meshtron,
title={Meshtron: High-Fidelity, Artist-Like 3D Mesh Generation at Scale},
author={Hao, Zekun and Romero, David W. and Lin, Tsung-Yi and Liu, Ming-Yu},
journal={arXiv preprint arXiv:2412.09548},
year={2024}
}