Unknown-Aware Diverse Prompt Learning for Open-Set Single Domain Generalization-Based Face Anti-spoofing

Fangling Jiang, Qi Li, Weining Wang, Bing Liu, Zhenan Sun

Abstract: Recently, the generalization capability of face anti-spoofing models has taken the attention of both industry and academia. Among all the problems, domain shift and unknown attacks are the most serious problems affecting generalization performance. Existing work usually focuses on dealing with one of the above problems. In this paper, we address a challenging but practical problem, open-set single domain generalization-based (OSSDG-based) face anti-spoofing, simultaneously addressing these two problems with limited training data. We propose a novel unknown-aware diverse prompt learning framework, which mines the visual-language pre-training knowledge for OSSDG-based face anti-spoofing by text prompt generation and unknown-aware learning regularization. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art performance on both known classes and unknown attacks in cross-scenario domains with a single training domain.

External IDs:dblp:conf/ccbr/JiangLWLS24