MeshMosaic: Scaling Artist Mesh Generation via Local-to-Global Assembly

2025-09-24

AI Summary

This work addresses the limitations of autoregressive grid generation models in modeling long sequences and achieving high quantization accuracy. To enable scalable generation of artist-designed meshes with over 100,000 triangles while preserving fine geometric details and structured density distributions, we propose a local-to-global block assembly framework. Our method comprises: (i) shape-aware mesh partitioning; (ii) autoregressive patch generation with shared boundary constraints; (iii) localized quantization; and (iv) Transformer architecture enhancements to ensure symmetry, coherence, and seamless stitching. Evaluated on multiple public datasets, our approach significantly improves geometric fidelity and user preference scores. It is the first method to achieve controllable, high-quality generation of high-resolution, high-complexity 3D meshes—establishing a new paradigm for precision-oriented 3D content creation.

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: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Scaling artist-designed meshes to high triangle numbers remains challenging for autoregressive generative models. Existing transformer-based methods suffer from long-sequence bottlenecks and limited quantization resolution, primarily due to the large number of tokens required and constrained quantization granularity. These issues prevent faithful reproduction of fine geometric details and structured density patterns. We introduce MeshMosaic, a novel local-to-global framework for artist mesh generation that scales to over 100K triangles--substantially surpassing prior methods, which typically handle only around 8K faces. MeshMosaic first segments shapes into patches, generating each patch autoregressively and leveraging shared boundary conditions to promote coherence, symmetry, and seamless connectivity between neighboring regions. This strategy enhances scalability to high-resolution meshes by quantizing patches individually, resulting in more symmetrical and organized mesh density and structure. Extensive experiments across multiple public datasets demonstrate that MeshMosaic significantly outperforms state-of-the-art methods in both geometric fidelity and user preference, supporting superior detail representation and practical mesh generation for real-world applications.

Problem

• Scaling artist-designed meshes to high triangle numbers challenges autoregressive models
• Transformer methods face long-sequence bottlenecks and limited quantization resolution
• Existing approaches struggle reproducing fine geometric details and structured density patterns

Innovation

• Local-to-global patch assembly framework
• Autoregressive patch generation with shared boundaries
• Individual patch quantization for scalability