VertexRegen: Mesh Generation with Continuous Level of Detail

ICCV 2025
Xiang Zhang1,*, Yawar Siddiqui2, Armen Avetisyan2, Chris Xie2, Jakob Engel2, Henry Howard-Jenkins2
* Work conducted while the author was an intern at Meta.
1UC San Diego 2Meta Reality Labs Research
Paper arXiv

Generation Demo

Generation Steps: /

Teaser: VertexRegen vs. Other Methods

VertexRegen generates mesh from coarse-to-fine with increasing level of detail.

Abstract

We introduce VertexRegen, a novel mesh generation framework that enables generation at a continuous level of detail. Existing autoregressive methods generate meshes in a partial-to-complete manner and thus intermediate steps of generation represent incomplete structures. VertexRegen takes inspiration from progressive meshes and reformulates the process as the reversal of edge collapse, i.e. vertex split, learned through a generative model. Experimental results demonstrate that VertexRegen produces meshes of comparable quality to state-of-the-art methods while uniquely offering anytime generation with the flexibility to halt at any step to yield valid meshes with varying levels of detail.

Method Overview

VertexRegen leverages edge collapse to generate training data, where the generation process can be achieved by reversing the edge collapse operation, i.e. vertex split.

Quantitative Evaluations

Quantitative comparisons with state-of-the-art methods

VertexRegen achieves comparable quality to state-of-the-art methods.

Quantitative evaluations under face constraints

Unconditional generation under face count constraints. VertexRegen achieves significantly better COV, MMD, and 1-NNA in early stages of generation.

BibTeX


  @article{zhang2025vertexregen,
    title={VertexRegen: Mesh Generation with Continuous Level of Detail},
    author={Zhang, Xiang and Siddiqui, Yawar and Avetisyan, Armen and Xie, Chris and Engel, Jakob and Howard-Jenkins, Henry},
    journal={arXiv preprint arXiv:2508.09062},
    year={2025}
  }