TY - GEN
T1 - 3D Face Recognition with Contrastive Learning Network on Low-Quality Data
AU - Jing, Yaping
AU - Mian, Ajmal
AU - Zhang, Leo
AU - Gao, Shang
AU - Lu, Xuequan
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
PY - 2025/2/27
Y1 - 2025/2/27
N2 - Face recognition has gained widespread use as a biometric technology. While many deep learning-based 3D face recognition techniques have achieved promising results using high-quality databases, recognizing faces on low-quality face data, often characterized by poses, occlusions, and temporal changes, remains a challenge, especially when captured with low-cost sensors. In this paper, we propose a novel end-to-end dual network using contrastive learning for 3D face recognition on low-quality data. In particular, we construct a pair of contrastive encoders with the MobileNet V2 backbone for contrastive representation learning. Furthermore, we introduce a joint loss function that combines the contrastive loss and the cross-entropy loss to facilitate joint contrastive learning and classification. Experiments show that our approach achieves state-of-the-art performance under different settings.
AB - Face recognition has gained widespread use as a biometric technology. While many deep learning-based 3D face recognition techniques have achieved promising results using high-quality databases, recognizing faces on low-quality face data, often characterized by poses, occlusions, and temporal changes, remains a challenge, especially when captured with low-cost sensors. In this paper, we propose a novel end-to-end dual network using contrastive learning for 3D face recognition on low-quality data. In particular, we construct a pair of contrastive encoders with the MobileNet V2 backbone for contrastive representation learning. Furthermore, we introduce a joint loss function that combines the contrastive loss and the cross-entropy loss to facilitate joint contrastive learning and classification. Experiments show that our approach achieves state-of-the-art performance under different settings.
KW - 3D face recognition
KW - contrastive learning
KW - low-quality 3D face data
UR - http://www.scopus.com/inward/record.url?scp=86000471676&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-81806-6_11
DO - 10.1007/978-3-031-81806-6_11
M3 - Conference paper
AN - SCOPUS:86000471676
SN - 9783031818059
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 145
EP - 157
BT - Advances in Computer Graphics - 41st Computer Graphics International Conference, CGI 2024, Proceedings
A2 - Magnenat-Thalmann, Nadia
A2 - Kim, Jinman
A2 - Sheng, Bin
A2 - Deng, Zhigang
A2 - Thalmann, Daniel
A2 - Li, Ping
PB - Springer Science + Business Media
CY - Switzerland
T2 - 41st Computer Graphics International Conference, CGI 2024
Y2 - 1 July 2024 through 5 July 2024
ER -