Few-data guided learning upon end-to-end point cloud network for 3D face recognition

Deep-learning-based 3D face recognition methods have developed vigorously in recent years, while the potential of these is being exploited more and scenarios. In this paper, an end-to-end deep learning network entitled Sur3dNet-Face for point-cloud-based proposed. The method uses PointNet, which a successful point cloud classification solution but performs unexpectedly recognition, as backbone. To adapt backbone to modifications architecture few-data guided framework based on Gaussian process morphable model supplemented. Instead mass data multiple datasets training, our takes only Spring2003 subset FRGC v2.0 training contains 943 facial scans well trained with such small amount real data. processing time generate representation less than 0.15 s. Without fine-tuning test set, Rank-1 Recognition Rate (RR1) achieved follows: 98.85% dataset 99.33% Bosphorus dataset, proves effectiveness potentiality method. When facing scenarios limited resource, proposed expected give competitive performance.