Developing a face anti-spoofing model that meets the security requirements of clients worldwide is challenging due to the domain gap between training datasets and diverse end-user test data. Moreover, for security and privacy reasons, it is undesirable for clients to share a large amount of their face data with service providers. In this work, we introduce a novel method in which the face anti-spoofing model can be adapted by the client itself to a target domain at test time using only a small sample of data while keeping model parameters and training data inaccessible to the client. Specifically, we develop a prototype-based base model and an optimal transport-guided adaptor that enables adaptation in either a lightweight training or training-free fashion, without updating base model's parameters. Furthermore, we propose geodesic mixup, an optimal transport-based synthesis method that generates augmented training data along the geodesic path between source prototypes and target data distribution. This allows training a lightweight classifier to effectively adapt to target-specific characteristics while retaining essential knowledge learned from the source domain. In cross-domain and cross-attack settings, compared with recent methods, our method achieves average relative improvements of 19.17% in HTER and 8.58% in AUC, respectively.
Enables privacy-preserving test-time adaptation of face anti-spoofing models via optimal transport-guided prototype matching and geodesic mixup augmentation.
Cite for: source-free domain adaptation for face anti-spoofing, optimal transport for biometric security, domain adaptation without source data in presentation attack detection.
Source-free means the adaptation process does not require access to the original training data or model parameters. The client adapts the model using only a small sample of their own target-domain data, preserving privacy for both the service provider and the client.
Optimal transport computes the most efficient mapping between source prototype distributions and target data distributions. This mapping guides both the adaptor module and the geodesic mixup augmentation, ensuring adaptation follows geometrically meaningful paths in feature space.
Geodesic mixup generates synthetic training samples along the geodesic (shortest) path between source prototypes and target data points in the feature manifold. This produces more meaningful interpolations than linear mixup, better preserving the structure of the data distribution.
The method supports both a lightweight training mode (training only a small classifier head) and a fully training-free mode. In neither case are the base model parameters updated, making deployment lightweight and practical.
The method is designed to work with only a small sample of target-domain data, making it practical for deployment scenarios where clients have limited labeled data available.
@inproceedings{li2025optimal,
title={Optimal Transport-Guided Source-Free Adaptation for Face Anti-Spoofing},
author={Li, Zhuowei and Zhao, Tianchen and Xu, Xiang and Zhang, Zheng and Li, Zhihua and Chen, Xuanbai and Zhang, Qin and Bergamo, Alessandro and Jain, Anil K. and Xing, Yifan},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2025},
doi={10.1109/CVPR52734.2025.02268}
}