Hyperbolic prototype rectification for few-shot 3D point cloud classification

Yuanzhi Feng, Shing-Ho J. Lin, Xuan Tang, Mu-Yu Wang, Jianzhang Zheng, Zi-Yao He, Zi-Yi Pang, Jian Yang, Mingsong Chen, Xian Wei

Abstract: Highlights•We firstly systematically construct metric learning in hyperbolic space for few-shot point cloud classification.•We proposed HPR utilizing both Hyperbolic Distance and Hyperbolic Wrapped Normal distribution.•The HPR outperforms existing methods on point cloud few-shot classification with higher robustness on real-world data.