第二届身份证呈现攻击检测竞赛
计算机视觉与模式识别
2025-07-29 v1
摘要
本工作总结并报告了第二届身份证呈现攻击检测竞赛的结果。与上一届相比,本届包含了一些新元素。(1) 启用了自动评估平台以进行自动基准测试;(2) 提出了两个赛道,分别用于评估算法和数据集;(3) 向赛道1的团队分享了新的身份证数据集,作为训练和优化的基线数据集。Hochschule Darmstadt、Fraunhofer-IGD和Facephi公司联合组织了这项挑战赛。共有20个团队注册,评估了74个提交的模型。对于赛道1,"Dragons"团队以40.48%的平均排名(AV-Rank)和11.44%的等错误率(EER)获得第一名。对于赛道2中更具挑战性的方法,"Incode"团队以14.76%的AV-Rank和6.36%的EER取得了最佳结果,相比第一届的74.30%和21.87%的EER有所提升。这些结果表明,身份证上的PAD正在改进,但它仍然是一个与图像数量,特别是真实图像数量相关的具有挑战性的问题。
引用
@article{arxiv.2507.20404,
title = {Second Competition on Presentation Attack Detection on ID Card},
author = {Juan E. Tapia and Mario Nieto and Juan M. Espin and Alvaro S. Rocamora and Javier Barrachina and Naser Damer and Christoph Busch and Marija Ivanovska and Leon Todorov and Renat Khizbullin and Lazar Lazarevich and Aleksei Grishin and Daniel Schulz and Sebastian Gonzalez and Amir Mohammadi and Ketan Kotwal and Sebastien Marcel and Raghavendra Mudgalgundurao and Kiran Raja and Patrick Schuch and Sushrut Patwardhan and Raghavendra Ramachandra and Pedro Couto Pereira and Joao Ribeiro Pinto and Mariana Xavier and Andrés Valenzuela and Rodrigo Lara and Borut Batagelj and Marko Peterlin and Peter Peer and Ajnas Muhammed and Diogo Nunes and Nuno Gonçalves},
journal= {arXiv preprint arXiv:2507.20404},
year = {2025}
}