SubdivAR: Autoregressive Next-Scale Prediction for Neural Mesh Subdivision

2026-06-25

AI Summary

This work addresses the limitations of traditional mesh subdivision methods, which suffer from oversmoothing due to fixed local rules, and existing neural approaches, which exhibit limited generalization owing to purely local modeling. To overcome these issues, we propose the Mesh Autoregressive Representation (MAR) framework, which reformulates subdivision as an autoregressive task of predicting vertex offsets at the next scale. By modeling an ordered sequence of scales, MAR effectively integrates global semantic information with local topological constraints. We introduce a hybrid topology-aware Transformer architecture tailored for this task and construct FII-40K, a high-quality multi-scale supervised dataset comprising 40,000 meshes. Experiments demonstrate that our method surpasses state-of-the-art approaches by 18.8% in Hausdorff Distance and 14.2% in Chamfer Distance, while exhibiting exceptional robustness on complex open surfaces.

Abstract

Mesh subdivision is a fundamental operation for converting coarse, editable meshes into high-resolution surfaces, with broad applications in digital asset creation. Classical rule-based schemes rely on fixed local refinement rules and often produce over-smoothed surfaces. Recent neural subdivision methods improve detail synthesis, but remain constrained by local modeling and exhibit limited generalizability. We present SubdivAR, a neural mesh subdivision framework based on our proposed Mesh Autoregressive Representation (MAR). MAR arranges meshes at different subdivision levels into an ordered scale sequence, reformulating subdivision as autoregressive next-scale prediction. To support this formulation, we introduce a Hybrid Topology-Aware Transformer that combines global semantic attention with topology-constrained local feature aggregation. SubdivAR adopts a next-scale coordinate prediction paradigm, regressing vertex offsets at each refinement stage to preserve subdivision topology while recovering fine-grained geometric details. To enable reliable learning, we construct FII-40K, a curated dataset of nearly 40,000 high-quality meshes with multi-level subdivision supervision. Experiments show that SubdivAR outperforms state-of-the-art baselines, reducing Hausdorff Distance and Chamfer Distance by 18.8% and 14.2%, respectively, and demonstrates strong robustness on complex open-surface geometries.

Problem

• mesh subdivision
• detail synthesis
• generalizability
• over-smoothed surfaces
• local modeling

Innovation

• autoregressive mesh representation
• neural subdivision
• Hybrid Topology-Aware Transformer
• next-scale prediction
• multi-level subdivision