yxzhang15 / PCR

A Collection of DL-based Point Cloud Registration Methods

This repository provides a summary of deep learning based point cloud registration algorithms.

If you find this repository helpful, we would greatly appreciate it if you could cite our paper: http://arxiv.org/abs/2404.13830.

We classify registration algorithms into supervised and unsupervised, as follows.

Supervised Point Cloud Registration Methods

• 1. Descriptor Extraction
  • 1. 1 Two-view
  • 1. 2 Multi-view
• 2. Correspondence Search
  • 2. 1 Partial-object
  • 2. 2 Full-object
• 3. Outlier Filtering
• 4. Transformation Parameter Estimation
• 5. Optimization
  • 5. 1 ICP-based
  • 5. 2 Probabilistic-based
• 6. Multimodal

Unsupervised Point Cloud Registration Methods

• 1. Correspondence-free
• 2. Correspondence-based

Datasets

Supervised Point Cloud Registration Methods

1. Descriptor Extraction

1. 1 Keypoint-based

• DeepVCP: An End-to-End Deep Neural Network for Point Cloud Registration. [paper] [code]

• The Perfect Match: 3D Point Cloud Matching With Smoothed Densities. [paper] [code]

• 3D Local Features for Direct Pairwise Registration. [paper] [code]

• HRegNet: A Hierarchical Network for Large-scale Outdoor LiDAR Point Cloud Registration. [paper] [code]

• SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration. [paper] [code]

• StickyPillars: Robust and Efficient Feature Matching on Point Clouds using Graph Neural Networks. [paper] [code]

• You Only Hypothesize Once: Point Cloud Registration with Rotation-equivariant Descriptors. [paper] [code]

• RoReg: Pairwise Point Cloud Registration with Oriented Descriptors and Local Rotations. [paper] [code]

• BUFFER: Balancing Accuracy, Efficiency, and Generalizability in Point Cloud Registration. [paper] [code]

• CoFiNet: Reliable Coarse-to-fine Correspondences for Robust PointCloud Registration. [paper] [code]

• One-Inlier is First: Towards Efficient Position Encoding for Point Cloud Registration. [paper] [code]

• GeDi: Learning General and Distinctive 3D Local Deep Descriptors for Point Cloud Registration. [paper] [code]

• GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer. [paper] [code]

• RoITr: Rotation-Invariant Transformer for Point Cloud Matching. [paper] [code]

1. 2 Multiview

• Learning and Matching Multi-View Descriptors for Registration of Point Clouds. [paper] [code]

• End-to-end learning local multi-view descriptors for 3d point clouds [paper] [code]

• Learning multiview 3d point cloud registration [paper] [code]

• Robust Multiview Point Cloud Registration with Reliable Pose Graph Initialization and History Reweighting. [paper] [code]

3. Correspondence Search

3. 1 Partial-object

• PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet. [paper] [code]

• Deep Closest Point: Learning Representations for Point Cloud Registration. [paper] [code]

• PointNetLK Revisited. [paper] [code]

3. 2 Partial-object

• PRNet: Self-supervised Learning for Partial-to-partial Registration. [paper] [code]

• RPM-Net: Robust Point Matching Using Learned Features. [paper] [code]

• PREDATOR: Registration of 3D Point Clouds with Low Overlap. [paper] [code]

• OMNet: Learning Overlapping Mask for Partial-to-Partial Point Cloud Registration. [paper] [code]

• STORM: Structure-Based Overlap Matching for Partial Point Cloud Registration. [paper] [code]

• FIRE-Net: Feature Interactive Representation for Point Cloud Registration. [paper] [code]

• PCAM: Product of Cross-Attention Matrices for Rigid Registration of Point Clouds. [paper] [code]

• REGTR: End-to-end Point Cloud Correspondences with Transformers. [paper] [code]

3. Outlier Filtering

• 3DRegNet: A Deep Neural Network for 3D Point Registration. [paper] [code]

• PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency. [paper] [code]

• DLF: Partial Point Cloud Registration With Deep Local Feature. [paper] [code]

• 3D Registration with Maximal Cliques. [paper] [code]

• Deep Hough Voting for Robust Global Registration. [paper] [code]

4. Transformation Parameter Estimation

• DeTarNet: Decoupling Translation and Rotation by Siamese Network for Point Cloud Registration. [paper] [code]

• FINet: Dual Branches Feature Interaction for Partial-to-Partial Point Cloud Registration [paper] [code]

5. Optimization

5. 1 ICP-based

• Deep Closest Point: Learning Representations for Point Cloud Registration. [paper] [code]

• PRNet: Self-supervised Learning for Partial-to-partial Registration. [paper] [code]

• IDAM: Iterative Distance-Aware Similarity Matrix Convolution with Mutual-Supervised Point Elimination for Efficient Point Cloud Registration. [paper] [code]

5. 2 Probabilistic-based

• DeepGMR: Learning Latent Gaussian Mixture Models for Registration. [paper] [code]

• OGMM: Overlap-guided Gaussian Mixture Models for Point Cloud Registration. [paper] [code]

• Point Cloud Registration Based on Learning Gaussian Mixture Models With Global-Weighted Local Representations. [paper] [code]

• VBReg: Robust Outlier Rejection for 3D Registration with Variational Bayes. [paper] [code]

6. Multimodal

• PCR-CG: Point Cloud Registration via Deep Explicit Color and Geometry. [paper] [code]

• PEAL: Prior-embedded Explicit Attention Learning for Low-overlap Point Cloud Registration. [paper] [code]

• ImLoveNet: Misaligned Image-supported Registration Network for Low-overlap Point Cloud Pairs. [paper] [code]

• IMFNet: Interpretable Multimodal Fusion for Point Cloud Registration. [paper] [code]

• GMF: General Multimodal Fusion Framework for Correspondence Outlier Rejection. [paper] [code]

Unsupervised Point Cloud Registration Methods

1. Correspondence-free

• PPF-FoldNet: Unsupervised Learning of Rotation Invariant 3D Local Descriptors. [paper] [code]

• PCRNet: Point Cloud Registration Network using PointNet Encoding [paper] [code]

• UPCR: A Representation Separation Perspective to Correspondence-Free Unsupervised 3-D Point Cloud Registration. [paper] [code]

• UGMM: Unsupervised Point Cloud Registration by Learning Unified Gaussian Mixture Models. [paper] [code]

• Research and Application on Cross-source Point Cloud Registration Method Based on Unsupervised Learning。 [paper] [code]

2. Correspondence-based

• CEMNet: Sampling Network Guided Cross-Entropy Method for Unsupervised Point Cloud Registration. [paper] [code]

• RIENet: Reliable Inlier Evaluation for Unsupervised Point Cloud Registration. [paper] [code]

• UDPReg: Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration. [paper] [code]

Datasets

• Stanford: A Volumetric Method for Building Complex Models from Range Images. [paper] [code]

• ETH: Challenging Data Sets for Point Cloud Registration Algorithms. [paper] [code]

• KITTI: Are We Ready for Autonomous Driving? The KITTI Vision Benchmark Suite. [paper] [code]

• ModelNet40: 3D ShapeNets: A Deep Representation for Volumetric Shapes. [paper] [code]

• ShapeNet: An Information-Rich 3D Model Repository. [paper] [code]

• ICL-NUIM: A Benchmark for RGB-D Visual Odometry, 3D Reconstruction and SLAM. [paper] [code]

• 3DMatch: Learning the Matching of Local 3D Geometry in Range Scans. [paper] [code]

• Apollo-SouthBay: L3-Net: Towards Learning Based LiDAR Localization for Autonomous Driving. [paper] [code]

• ScanObjectNN. Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data. [paper] [code]

• WHU-TLS: Registration of Large-scale Terrestrial Laser Scanner Point Clouds: A Review and Benchmark. [paper] [code]

• FlyingShapes: SIRA-PCR: Sim-to-Real Adaptation for 3D Point Cloud Registration. [paper] [code]