SubdivAR: Autoregressive Next-Scale Prediction for Neural Mesh Subdivision

2026-06-25 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, quick

Summary

SubdivAR is a novel neural mesh subdivision framework designed to convert coarse, editable meshes into high-resolution surfaces, addressing limitations of classical rule-based and prior neural methods. It introduces the Mesh Autoregressive Representation (MAR), which organizes meshes across different subdivision levels into an ordered scale sequence, reframing subdivision as an autoregressive next-scale prediction task. To facilitate this, SubdivAR employs a Hybrid Topology-Aware Transformer that integrates global semantic attention with topology-constrained local feature aggregation. The framework utilizes a next-scale coordinate prediction paradigm, regressing vertex offsets at each refinement stage to maintain subdivision topology while accurately recovering fine geometric details. Trained on FII-40K, a curated dataset of nearly 40,000 high-quality meshes, SubdivAR significantly outperforms state-of-the-art baselines, reducing Hausdorff Distance by 18.8% and Chamfer Distance by 14.2%, and demonstrates strong robustness on complex open-surface geometries.

Key takeaway

For AI Scientists developing high-resolution 3D asset creation tools, SubdivAR offers a superior neural mesh subdivision approach. You should consider its autoregressive next-scale prediction paradigm and Hybrid Topology-Aware Transformer to overcome limitations of classical and prior neural methods, achieving better detail synthesis and robustness on complex geometries. This could significantly enhance the quality and efficiency of your digital asset pipelines.

Key insights

SubdivAR reformulates mesh subdivision as autoregressive next-scale prediction using a novel Mesh Autoregressive Representation.

Principles

• Mesh subdivision can be modeled as an ordered scale sequence.
• Combining global semantic attention with local topology improves neural subdivision.
• Regressing vertex offsets preserves topology while adding detail.

Method

SubdivAR uses MAR to arrange meshes into scale sequences, then employs a Hybrid Topology-Aware Transformer for autoregressive next-scale coordinate prediction, regressing vertex offsets.

In practice

• Use autoregressive prediction for multi-level mesh refinement.
• Integrate global and local features in mesh processing.
• Leverage large, curated datasets like FII-40K for training.

Topics

• Mesh Subdivision
• Neural Networks
• 3D Modeling
• Autoregressive Models
• Transformers
• Geometric Deep Learning
• FII-40K Dataset

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• S23DR 2026: End-to-End 3D Wireframe Prediction via DETR-Style Set Prediction with Contrastive Denoising — WireframeDETR directly predicts 3D wireframes from point clouds using DETR-style set prediction, enhanced by contrastive denoising and multi-scale encoding.
• CamDirector: Towards Long-Term Coherent Video Trajectory Editing — CamDirector enhances video trajectory editing with precise camera control and long-term consistency via hybrid warping and history-guided autoregressive diffusion.
• Adaptive Volumetric Mechanical Property Fields Invariant to Resolution — AdaVoMP predicts dense, spatially-varying mechanical properties for 3D objects using an adaptive sparse voxel structure and a transformer model.
• RigPAPR: Rig-Based Animation of Static Neural Point Clouds from a Fixed-Viewpoint Video — Interpolation-based neural point rendering inherently avoids joint-boundary artifacts in rigged animation, unlike fixed-shape primitives.
• Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective — Feed-forward 3D reconstruction offers efficient, generalizable scene modeling by directly mapping 2D inputs to 3D representations.
• On Geometry Regularization in Autoencoder Reduced-Order Models with Latent Neural ODE Dynamics — Stiefel projection effectively regularizes autoencoder latent spaces for improved neural ODE dynamics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.