Unified Physical-Digital Attack Detection Challenge

TL;DR

This work tackles the need for unified detection of physical and digital face attacks in face recognition systems by introducing UniAttackData, the largest public dataset capable of representing both attack types on a per-identity basis. Built around two evaluation protocols, including leave-one-type-out generalization, the study reports on a CVPR2024 challenge with 133 registrations and 13 finalists, analyzing how participants fuse data augmentation, cross-domain generalization, and modern architectures to detect unified attacks. Key contributions include a public, scalable dataset, a unified evaluation framework, and detailed analyses of top-performing methods, highlighting practical design choices such as transformer backbones, 3D representations, and targeted augmentations that improve cross-threat performance. The work provides actionable directions for future UAD research, including leveraging large vision-language models, expanding attack types, and refining protocols to better simulate real-world, hybrid threat landscapes.

Abstract

Face Anti-Spoofing (FAS) is crucial to safeguard Face Recognition (FR) Systems. In real-world scenarios, FRs are confronted with both physical and digital attacks. However, existing algorithms often address only one type of attack at a time, which poses significant limitations in real-world scenarios where FR systems face hybrid physical-digital threats. To facilitate the research of Unified Attack Detection (UAD) algorithms, a large-scale UniAttackData dataset has been collected. UniAttackData is the largest public dataset for Unified Attack Detection, with a total of 28,706 videos, where each unique identity encompasses all advanced attack types. Based on this dataset, we organized a Unified Physical-Digital Face Attack Detection Challenge to boost the research in Unified Attack Detections. It attracted 136 teams for the development phase, with 13 qualifying for the final round. The results re-verified by the organizing team were used for the final ranking. This paper comprehensively reviews the challenge, detailing the dataset introduction, protocol definition, evaluation criteria, and a summary of published results. Finally, we focus on the detailed analysis of the highest-performing algorithms and offer potential directions for unified physical-digital attack detection inspired by this competition. Challenge Website: https://sites.google.com/view/face-anti-spoofing-challenge/welcome/challengecvpr2024.

Figures (10)

• Figure 1: The overall pipeline of MTFace's method. The moiré augmentation improves the model’s PAD ability, while self-blending augmentation improves the model’s DAD ability.
• Figure 2: The pipeline of Team duileduile's framework with pre-training stage and fine-tuning stage.
• Figure 3: This figure shows the DBPixBiS's architecture, the feature map, and the final outputs are supervised by angular loss.
• Figure 4: The overall structure of team VAI-Face's solution.
• Figure 5: The overall framework of team M2-Purdue's algorithm.