MeshLLM: Empowering Large Language Models to Progressively Understand and Generate 3D Mesh

TL;DR

MeshLLM introduces a large-scale dataset and novel strategies enabling large language models to better understand and generate 3D meshes, significantly improving performance over previous methods.

Contribution

The paper presents a new framework with a Primitive-Mesh decomposition and local mesh assembly to enhance LLMs' ability to process 3D mesh data.

Findings

Outperforms state-of-the-art in mesh generation quality

Creates a dataset 50 times larger than previous methods

Enhances LLMs' understanding of 3D shape topology

Abstract

We present MeshLLM, a novel framework that leverages large language models (LLMs) to understand and generate text-serialized 3D meshes. Our approach addresses key limitations in existing methods, including the limited dataset scale when catering to LLMs' token length and the loss of 3D structural information during mesh serialization. We introduce a Primitive-Mesh decomposition strategy, which divides 3D meshes into structurally meaningful subunits. This enables the creation of a large-scale dataset with 1500k+ samples, almost 50 times larger than previous methods, which aligns better with the LLM scaling law principles. Furthermore, we propose inferring face connectivity from vertices and local mesh assembly training strategies, significantly enhancing the LLMs' ability to capture mesh topology and spatial structures. Experiments show that MeshLLM outperforms the state-of-the-art LLaMA-Mesh in both mesh generation quality and shape understanding, highlighting its great potential in processing text-serialized 3D meshes.