Learned Radius Estimation for UDF-Based Point Cloud Reconstruction

2026-06-17 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

Surface reconstruction from point clouds is crucial for consumer-grade 3D capture applications like AR/VR and indoor scanning. Local-patch Unsigned Distance Field (UDF) methods offer lightweight and generalizable solutions, but their accuracy is highly dependent on the support radius. Traditionally, this radius is fixed or chosen via a one-dimensional curvature heuristic, which fails to capture heterogeneous local geometry effectively. A new approach proposes a learned per-query radius selector that predicts a continuous support radius. This selector plugs into a frozen LoSF-UDF backbone and is trained using off-grid target radii, obtained through parabolic interpolation of cached UDF error curves. Experiments demonstrate that this method significantly improves fine-scale reconstruction accuracy.

Key takeaway

For Computer Vision Engineers or AR/VR developers focused on improving 3D capture accuracy, you should investigate integrating learned radius estimation into your UDF-based point cloud reconstruction pipelines. This method directly addresses the limitations of fixed or heuristic support radii, enabling significantly improved fine-scale reconstruction for heterogeneous local geometries in applications like indoor scanning. Consider this approach to achieve more precise and detailed surface models.

Key insights

Learning to adapt the support radius dynamically improves UDF-based point cloud reconstruction accuracy, especially for complex geometries.

Principles

• UDF accuracy depends on support radius.
• Fixed radii limit reconstruction quality.
• Learned adaptation enhances geometric detail.

Method

A learned per-query radius selector predicts a continuous support radius, integrating into a frozen LoSF-UDF backbone. It trains on off-grid target radii derived from parabolic interpolation of cached UDF error curves.

In practice

• Improve AR/VR 3D capture.
• Enhance indoor scanning detail.
• Optimize UDF for varied geometries.

Topics

• Point Cloud Reconstruction
• Unsigned Distance Fields
• Learned Radius Estimation
• Computer Vision
• AR/VR
• 3D Capture

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

Related on AIssential

• Zero-Shot Polygon Matching with Pre-trained Models for Pose Estimation and Polygon Cloud from Challenging Stereo — Unsupervised polygon matching for stereo images is achievable by integrating pre-trained models with handcrafted features.
• FlashFPS: Efficient Farthest Point Sampling for Large-Scale Point Clouds via Pruning and Caching — FlashFPS accelerates point cloud processing by pruning and caching Farthest Point Sampling computations.
• 3D Consistency Optimization for Self-Supervised Monocular Video Depth Estimation — Recasting monocular video depth as multi-view 3D reconstruction with 3D consistency optimization improves geometric accuracy and coherence.
• Breaking the Rigid Prior: Towards Articulated 3D Anomaly Detection — Articulated 3D anomaly detection requires disentangling pose-induced geometry from true structural defects.
• Towards Learning a Generalizable 3D Scene Representation from 2D Observations — A Generalizable Neural Radiance Field predicts 3D workspace occupancy from robot observations in a global frame.
• Modeling Depth Ambiguity: A Mixture-Density Representation for Flying-Point-Free Depth Estimation — Modeling depth ambiguity with multiple hypotheses per pixel prevents flying points and improves boundary reconstruction in depth estimation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.