RegiFormer: Unsupervised Point Cloud Registration via Geometric Local-to-Global Transformer and Self-Augmentation
- 29 July 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-13
Abstract
Representation learning on two but partial point clouds remains an open challenge in unsupervised point cloud registration (U-PCR). In this paper, we introduce RegiFormer, a geometric local-to-global transformer-based unsupervised framework equipped with a self-augmentation strategy, for point cloud registration. The geometric local-to-global transformer aggregates features not only from local neighborhoods but also extracts global intra-relationships within the entire point cloud using a transformation-invariant geometry embedding. Additionally, it enhances the inter-relationships between paired point clouds. To overcome the limited ability of U-PCR methods to learn alignment knowledge, we design a self-augmentation strategy that can be flexibly integrated into advanced models, significantly boosting their registration performance. Extensive experiments, conducted on five popular synthetic and real-scanned benchmarks, demonstrate the superior performance of RegiFormer when compared to state-of-the-art methods, both qualitatively and quantitatively.Keywords
Funding Information
- Research Grants Council of the Hong Kong Special Administrative Region, China (UGC/FDS16/E14/21)
- National Natural Science Foundation of China (T2322012, 62172218, 62032011)
- Shenzhen Science and Technology Program (JCYJ20220818103401003, JCYJ20220530172403007)
- Basic and Applied Basic Research Foundation of Guangdong Province (2022A1515010170)
This publication has 63 references indexed in Scilit:
- Point TransformerIEEE Access, 2021
- Iterative Distance-Aware Similarity Matrix Convolution with Mutual-Supervised Point Elimination for Efficient Point Cloud Registration2020
- DeepGMR: Learning Latent Gaussian Mixture Models for Registration2020
- Info3D: Representation Learning on 3D Objects Using Mutual Information Maximization and Contrastive Learning2020
- 3D ShapeNets: A deep representation for volumetric shapes2015
- Scene Coordinate Regression Forests for Camera Relocalization in RGB-D Images2013
- Are we ready for autonomous driving? The KITTI vision benchmark suite2012
- Model globally, match locally: Efficient and robust 3D object recognition2010
- Global Optimization through Rotation Space SearchInternational Journal of Computer Vision, 2009
- A method for registration of 3-D shapesIEEE Transactions on Pattern Analysis and Machine Intelligence, 1992