CTV-FAS: Compensate Texts with Visuals for Generalizable Face Anti-spoofing
Keywords: Face anti-spoofing Vision-language model Domain generalization
Abstract: Generalizable Face Anti-Spoofing (FAS) approaches have recently gained significant attention for their robustness in unseen scenarios.
Recent methods incorporate vision-language models into FAS, capitalizing on their remarkable pre-trained performance to enhance generalization.
These methods predominantly rely on text prompts to learn the concept of attacks in FAS.
However, certain attacks, such as high-resolution replay attacks, cannot be described linguistically.
Relying solely on text prompts cannot accurately tackle such attacks, resulting in performance degradation.
To tackle these limitations, we introduce a novel framework named CTV-FAS, designed to exploit visual anchors to compensate for the shortcomings of semantic prompts.
Specifically, we employ a Self-Supervised Consistency Module (SSCM) to boost the generalization of visual anchors, which utilizes consistency regularization to facilitate visual feature learning.
Subsequently, a Visual Anchors Updating Module (VAUM) is proposed to incorporate the visual anchors through an adaptive updating scheme, guiding the feature learning process from a visual standpoint.
Furthermore, we propose an Adaptive Modality Integration Module (AMIM), designed to merge visual and textual information during inference seamlessly. This integration optimizes the synergy between modalities, significantly boosting the efficacy of Face Anti-Spoofing (FAS) tasks.
Our extensive experimental evaluations and in-depth analysis affirm that our method outperforms current state-of-the-art counterparts with a notable margin of superiority.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1824
Loading