UniAttackData+: Unified Physical-Digital Attack Detection+ Challenge

Hui Ma, Ajian Liu, Yongze Li, Chuanbiao Song, Xiao Guo, Changtao Miao, Wanyi Zhuang, Junze Zheng, Shunxin Chen, Yan Hong, Jiabao Guo, Jiankang Deng, Jun Lan, Weiqiang Wang, Tao Gong, Qi Chu, Sergio Escalera, Hugo Jair Escalante, Xiaoming Liu, Zhen Lei, Isabelle Guyon, Yanyan Liang, Jun Wan; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 3171-3180

Unified physical-digital face attack detection aims to develop a universal model capable of simultaneously detecting both digital and physical spoofing attacks. However, existing training datasets generally lack comprehensive coverage of diverse attack types, which limits the generalization ability of detection models and hinders their effectiveness in real-world applications. To address this limitation, we released a significantly expanded dataset, UniAttackData+, at the 6th Face Anti-Spoofing Workshop @ICCV 2025. The dataset includes 2,875 participants from three major demographic groups--Africa, East Asia, and Central Asia--and contains 18,250 real videos collected under diverse lighting conditions, backgrounds, and acquisition devices. For each participant, 54 types of attacks were simulated, comprising 14 physical and 40 digital attack variants, resulting in a total of 679,097 high-quality forged videos. Based on this dataset, we organized a Unified Attack Detection Challenge, which attracted 137 participating teams, with 12 teams advancing to the final round. The final rankings were determined based on results re-verified by the organizers. This paper reviews the challenge by introducing the dataset construction, protocol definitions, evaluation metrics, and competition results, and analyzes the top-performing algorithmic solutions, while also outlining future directions for unified physical-digital attack detection. Challenge Website: https://sites.google.com/view/face-anti-spoofing-challenge/welcome/challengeiccv2025?authuser=0