Meta-learning-based adversarial training for deep 3D face recognition on point clouds

作者：

Highlights：

• We propose an algorithm for deep 3D face recognition on point clouds, where an adversarial training based 3D face data augmentation and a meta-learning based network training are designed.

• Our algorithm generates adversarial 3D face samples for online and dynamic data augmentation.

• We propose a meta-learning framework to train the 3D face recognition model.

• Our algorithm achieves state-of-the-art results.

摘要

•We propose an algorithm for deep 3D face recognition on point clouds, where an adversarial training based 3D face data augmentation and a meta-learning based network training are designed.•Our algorithm generates adversarial 3D face samples for online and dynamic data augmentation.•We propose a meta-learning framework to train the 3D face recognition model.•Our algorithm achieves state-of-the-art results.

论文关键词：Deep 3D face recognition,Point clouds,Adversarial samples,Meta-learning

论文评审过程：Received 19 April 2022, Revised 18 June 2022, Accepted 20 September 2022, Available online 23 September 2022, Version of Record 6 October 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.109065