3D Focusing-and-Matching Network for Multi-Instance Point Cloud Registration (3DFMNet)
This research proposes a novel two-stage deep learning method, 3DFMNet, for multi-instance point cloud registration, which achieves state-of-the-art performance by first focusing on individual object centers for proposal generation and then performing pairwise registration between the model and each proposal.
Micro-Structures Graph-Based Point Cloud Registration: Balancing Efficiency and Accuracy Using a Novel Two-Stage Method
This paper introduces a novel, efficient, and accurate two-stage point cloud registration method that leverages a micro-structures graph to achieve robust coarse registration and precise fine registration.
Memory-Efficient 3D Point Cloud Registration Using Overlapping Region Sampling and Deep Learning
This research paper introduces a novel sampling method for 3D point cloud registration that prioritizes overlapping regions to reduce GPU memory consumption while maintaining high registration accuracy, especially for large-scale point clouds.
A Consistency-Aware Spot-Guided Transformer for Accurate and Efficient Point Cloud Registration
This paper introduces CAST, a novel deep learning architecture for point cloud registration that leverages consistency-aware attention mechanisms and a sparse-to-dense fine-matching module to achieve state-of-the-art accuracy and efficiency without relying on computationally expensive methods like optimal transport.
SE(3) Equivariant Graph Network for Sparse Point Cloud Registration: An Efficient and Robust Approach
This paper introduces Equi-GSPR, a novel graph neural network model designed for sparse point cloud registration, which leverages SE(3) equivariance to achieve robust and efficient performance by capturing geometric topology and effectively mitigating outlier correspondences.
LoGDesc: A Hybrid Descriptor for Robust Point Cloud Registration Using Local Geometric Features and Learning-Based Feature Propagation
This paper introduces LoGDesc, a novel hybrid descriptor that leverages local geometric features and learning-based feature propagation to achieve robust 3D point cloud registration, particularly in challenging scenarios with noise and low overlap.
Robust Point Cloud Registration with Neural Diffusion and Transformer
A robust point cloud registration approach that leverages graph neural partial differential equations and heat kernel signatures to enhance the robustness of feature representations and efficiently obtain corresponding keypoints.
A Robust and Efficient Baseline for Point Cloud Registration via Direct Superpoint Matching
Our approach directly matches superpoints between input point clouds to robustly estimate the SE(3) transformation matrix, without relying on cumbersome post-processing steps.
SPEAL: Skeletal Prior Embedded Attention Learning for Cross-Source Point Cloud Registration
Effektive Nutzung von Skelettgeometrie für präzise Punktewolkenregistrierung.
SPEAL: Skeletal Prior Embedded Attention Learning for Cross-Source Point Cloud Registration
SPEAL는 해부학적 선입견을 활용하여 점군의 지오메트리 복잡성을 캡처하고 등록을 용이하게 합니다.
© 2024 by Linnk AI