Partial Symmetry Detection for 3D Geometry using Contrastive Learning with Geodesic Point Cloud Patches

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

A novel method for partial extrinsic symmetry detection in 3D geometry is introduced, leveraging contrastive learning to generate rotation, reflection, translation, and scale-invariant local shape features from geodesic point cloud patches. The approach, implemented as self-supervised SymCL and supervised SymML models, can extract multiple valid symmetry solutions and demonstrates strong generalization across diverse object classes and datasets. The authors also present a new benchmark, SymPartNet.v1, for evaluating partial extrinsic symmetry detection, which includes Pre-Segmented Partial Symmetry Benchmark (PSPSB) and Partial Symmetry Benchmark (PSB). Quantitative evaluation on the "Chair" class of SymPartNet.v1 shows SymML slightly outperforms SymCL, with both models recovering 77.27% and 73.33% of ground truth symmetries respectively, achieving high IoU scores (76.63% and 77.83%) and low ICP distances (0.0028 and 0.0018). Ablation studies confirm the critical role of geodesic patches and data augmentation.

Key takeaway

For 3D computer vision engineers developing shape analysis or generation systems, this research offers a robust method for partial extrinsic symmetry detection. You should consider integrating self-supervised contrastive learning with geodesic patches to extract invariant features, improving generalization across diverse 3D models. This approach enables the identification of multiple valid symmetries and can enhance downstream tasks like segmentation or completion.

Key insights

A self-supervised contrastive learning approach using geodesic patches effectively detects partial extrinsic symmetries in 3D geometry.

Principles

• Geodesic patches improve local 3D geometry representation.
• Invariant features are crucial for robust symmetry detection.
• Self-supervised contrastive learning extracts transferable features.

Method

The pipeline involves feature learning via contrastive learning on geodesic patches, clustering features in latent space, and extracting symmetries, followed by a region growing algorithm for completion.

In practice

• Use geodesic patches for accurate local 3D geometry.
• Employ contrastive learning for self-supervised feature extraction.
• Apply region growing to maximize detected symmetric areas.

Topics

• Partial Symmetry Detection
• 3D Geometry Analysis
• Contrastive Learning
• Geodesic Point Clouds
• Self-Supervised Learning
• Deep Metric Learning

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

Related on AIssential

• Coarse-to-Fine: A Hybrid Self-Supervised Method for Non-rigid 3D Shape Matching — A hybrid self-supervised coarse-to-fine method enhances non-rigid 3D shape matching via dual functional map streams.
• Rotational Symmetry based Object Pose Estimation from Point Clouds in the Absence of Known 3D Models — Leveraging rotational symmetry enables robust object pose estimation from point clouds even without known 3D models, using iterative optimization.
• Learning Geometric Representations from Videos for Spatial Intelligent Multimodal Large Language Models — GeoVR enhances MLLMs' 3D spatial intelligence by distilling geometric knowledge from 2D videos using multi-objective learning.
• Topology-Aware Skeleton Detection via Lighthouse-Guided Structured Inference — Lighthouse-Skel proposed a dual-branch, topology-aware skeleton detection method using structured inference and a lighthouse-guided completion strategy.
• GAP-GDRNet: Geometry-Aware Monocular Visual Pose Sensing on a Single-Target Synthetic Spacecraft Dataset — GAP-GDRNet enhances 6D spacecraft pose sensing by integrating attention mechanisms for robust geometric feature extraction.
• GeoSAM-3D: Geodesic Prompt Propagation for Open-Vocabulary 3D Scene Segmentation from Monocular Video — GeoSAM-3D propagates 3D segmentation prompts using geodesic heat-kernel distance on Gaussian splatting scenes from monocular video, improving mask continuity and reducing leakage.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.