NFR: Neural Feature-Guided Non-Rigid Shape Registration

2025-05-28

AI Summary

This paper addresses unsupervised 3D shape registration under large-scale non-rigid deformations and partial overlap. The proposed method introduces a self-supervised framework that jointly leverages neural feature learning and geometric optimization. Its core innovation lies in the first integration of self-supervised deep feature embeddings into a differentiable iterative closest point (ICP) pipeline, augmented by a manifold-consistency-driven dynamic correspondence filtering mechanism to achieve synergistic semantic accuracy and geometric robustness. Crucially, the approach requires no ground-truth correspondences and enables end-to-end training using only dozens of low-diversity samples. Extensive evaluation on multiple non-rigid and partial-matching benchmarks demonstrates significant improvements over both classical and intrinsic methods. It particularly excels on unseen shape pairs exhibiting severe deformations, textureless surfaces, and weak structural cues—achieving state-of-the-art performance.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: 3D Computer Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

In this paper, we propose a novel learning-based framework for 3D shape registration, which overcomes the challenges of significant non-rigid deformation and partiality undergoing among input shapes, and, remarkably, requires no correspondence annotation during training. Our key insight is to incorporate neural features learned by deep learning-based shape matching networks into an iterative, geometric shape registration pipeline. The advantage of our approach is two-fold -- On one hand, neural features provide more accurate and semantically meaningful correspondence estimation than spatial features (e.g., coordinates), which is critical in the presence of large non-rigid deformations; On the other hand, the correspondences are dynamically updated according to the intermediate registrations and filtered by consistency prior, which prominently robustify the overall pipeline. Empirical results show that, with as few as dozens of training shapes of limited variability, our pipeline achieves state-of-the-art results on several benchmarks of non-rigid point cloud matching and partial shape matching across varying settings, but also delivers high-quality correspondences between unseen challenging shape pairs that undergo both significant extrinsic and intrinsic deformations, in which case neither traditional registration methods nor intrinsic methods work.

Problem

• Overcoming significant non-rigid deformation in 3D shape registration
• Addressing partiality in input shapes without correspondence annotation
• Improving correspondence accuracy using neural features in registration

Innovation

• Learning-based framework for 3D shape registration
• Neural features enhance correspondence estimation accuracy
• Dynamic correspondence updates robustify registration pipeline