CASIA-SURF CeFA: A Benchmark for Multi-modal Cross-ethnicity Face Anti-spoofing

Abstract

The availability of handy multi-modal (i.e., RGB-D) sensors has brought about a surge of face anti-spoofing research. However, the current multi-modal face presentation attack detection (PAD) has two defects: (1) The framework based on multimodal fusion requires providing modalities consistent with the training input, which seriously limits the deployment scenario.(2) The performance of ConvNet-based model on high fidelity datasets is increasingly limited. In this work, we present a pure transformer-based framework, dubbed the Flexible Modal Vision Transformer (FM-ViT), for face anti-spoofing to flexibly target any single-modal (i.e., RGB) attack scenarios with the help of available multi-modal data. Specifically, FM-ViT retains a specific branch for each modality to capture different modal information and introduces the Cross-Modal Transformer Block (CMTB), which consists of two cascaded attentions named Multi-headed Mutual-Attention (MMA) and Fusion-Attention (MFA) to guide each modal branch to mine potential features from informative patch tokens, and to learn modality-agnostic liveness features by enriching the modal information of own CLS token, respectively. Experiments demonstrate that the single model trained based on FM-ViT can not only flexibly evaluate different modal samples, but also outperforms existing single-modal frameworks by a large margin, and approaches the multi-modal frameworks introduced with smaller FLOPs and model parameters.

Adaptive Transformers for Robust Few-shot Cross-domain Face Anti-spoofing

“…Following the prior works, we regard each dataset as one domain and apply the leave-one-out testing protocol to evaluate the cross-domain generalization. In Protocol 2, we conduct similar cross-domain evaluations on the larger-scale datasets: CASIA-SURF [77,76], CASIA-CeFA [33,34], and WMCA [17]. Compared to datasets in Protocol 1, datasets in Protocol 2 have much more subjects and richer environment variations, and thus the results can better reflect model performance.…”

Cross‐ethnicity face anti‐spoofing recognition challenge: A review

“…CASIA‐SURF CeFA [22] is the largest up‐to‐date CeFA dataset, covering 3 ethnicities, 3 modalities, 1604 subjects, and 2D plus 3D attack types. More importantly, it is the first public dataset designed for exploring the impact of cross‐ethnicity in the study of face anti‐spoofing.…”

“…However, these algorithms are not accurate enough because of the use of handcrafted features, such as LBP [7–9], HoG [8–10] and GLCM [10], that do not necessarily capture the most discriminative information associated to the data. Recently, CNN‐based face PAD methods [11–16] have shown impressive progress due to the excellent performance of deep neural networks [11,14,15,17] and the availability of large datasets [15,18–22]. Although these methods achieve near‐perfect performance in intra‐database experiments, they are still vulnerable when facing complex authentication scenarios.…”

Online Adaptive Personalization for Face Anti-spoofing

“…While large face anti-spoofing datasets [6,12,13] have been collected in recent years to enable the development of deep learning solutions, it is infeasible to capture all the variations in the data that might appear at test time. Distribution shift between training and test data naturally occurs due to the presence of new users, sensors, and environmental conditions which were not captured in the training data.…”