GPI-Net: Gestalt-Guided Parallel Interaction Network via Orthogonal Geometric Consistency for Robust Point Cloud Registration

2025-07-18

AI Summary

To address the challenges of effectively fusing local and global features and robustly identifying high-quality correspondences in point cloud registration, this paper proposes a Gestalt-inspired parallel interaction network. Our method introduces three key innovations: (1) a Gestalt Feature Attention module that models structural completeness at the perceptual level; (2) a dual-path, multi-granularity parallel interaction architecture that jointly leverages self-attention and cross-attention, augmented by an orthogonal geometric consistency constraint to strengthen global structural representation; and (3) an orthogonal feature fusion strategy to enhance complementarity across granularities. Extensive experiments on standard benchmarks—including ModelNet40 and 3DMatch—demonstrate significant improvements over state-of-the-art methods in both matching accuracy and noise robustness. The source code is publicly available.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision
• Machine Learning: Learning with Manifolds

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
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

The accurate identification of high-quality correspondences is a prerequisite task in feature-based point cloud registration. However, it is extremely challenging to handle the fusion of local and global features due to feature redundancy and complex spatial relationships. Given that Gestalt principles provide key advantages in analyzing local and global relationships, we propose a novel Gestalt-guided Parallel Interaction Network via orthogonal geometric consistency (GPI-Net) in this paper. It utilizes Gestalt principles to facilitate complementary communication between local and global information. Specifically, we introduce an orthogonal integration strategy to optimally reduce redundant information and generate a more compact global structure for high-quality correspondences. To capture geometric features in correspondences, we leverage a Gestalt Feature Attention (GFA) block through a hybrid utilization of self-attention and cross-attention mechanisms. Furthermore, to facilitate the integration of local detail information into the global structure, we design an innovative Dual-path Multi-Granularity parallel interaction aggregation (DMG) block to promote information exchange across different granularities. Extensive experiments on various challenging tasks demonstrate the superior performance of our proposed GPI-Net in comparison to existing methods. The code will be released at https://github.com/gwk/GPI-Net.

Problem

• Improving point cloud registration via local-global feature fusion
• Reducing feature redundancy with orthogonal geometric consistency
• Enhancing correspondence quality using Gestalt-guided attention mechanisms

Innovation

• Gestalt-guided Parallel Interaction Network for registration
• Orthogonal integration strategy reduces redundant information
• Dual-path Multi-Granularity block enhances information exchange