Mesh Silksong: Auto-Regressive Mesh Generation as Weaving Silk

2025-07-03

AI Summary

Existing mesh tokenization methods suffer from sequence redundancy due to repeated vertex encoding, degrading generative efficiency and geometric integrity. To address this, we propose an autoregressive mesh generation paradigm inspired by the “silk-weaving” process: leveraging topology-guided single-pass vertex traversal, we construct a compact, repetition-free vertex sequence—achieving, for the first time, vertex-level unique encoding. Our method integrates geometry-aware tokenization, manifold-constrained modeling, and normal consistency optimization to effectively suppress non-manifold structures and holes. Experiments demonstrate that, compared to state-of-the-art approaches, our method reduces sequence redundancy by 50%, improves compression ratio by ~22%, and consistently outperforms prior work in watertightness, topological correctness, and geometric fidelity—enabling high-fidelity 3D content generation.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

We introduce Mesh Silksong, a compact and efficient mesh representation tailored to generate the polygon mesh in an auto-regressive manner akin to silk weaving. Existing mesh tokenization methods always produce token sequences with repeated vertex tokens, wasting the network capability. Therefore, our approach tokenizes mesh vertices by accessing each mesh vertice only once, reduces the token sequence's redundancy by 50%, and achieves a state-of-the-art compression rate of approximately 22%. Furthermore, Mesh Silksong produces polygon meshes with superior geometric properties, including manifold topology, watertight detection, and consistent face normals, which are critical for practical applications. Experimental results demonstrate the effectiveness of our approach, showcasing not only intricate mesh generation but also significantly improved geometric integrity.

Problem

• Auto-regressive mesh generation with reduced redundancy
• Improved mesh compression rate by 50%
• Ensuring superior geometric properties in meshes

Innovation

• Auto-regressive mesh generation like silk weaving
• Tokenizes vertices once, reducing redundancy by 50%
• Ensures manifold topology and watertight mesh properties