DFIC
收藏资源简介:
DFIC(Diverse Face Images - Coimbra)是由科英布拉大学·系统与机器人研究所构建的大规模面部图像数据集,旨在解决ICAO标准合规性验证的数据稀缺问题。该数据集包含58,633张高质量与低质量设备拍摄的图像及2,706段短视频,覆盖1,016名不同年龄、性别和种族的受试者,涵盖26项ICAO非合规场景(如遮挡、光照异常等),并附带90万手动标注和40万自动标注。数据通过多设备采集和受控条件模拟真实场景,应用于自动化护照照片合规检测、人脸识别系统公平性提升等领域。
DFIC (Diverse Face Images - Coimbra) is a large-scale facial image dataset developed by the Institute of Systems and Robotics, University of Coimbra, designed to address the data scarcity problem in ICAO standard compliance verification. This dataset contains 58,633 images captured by both high-quality and low-quality imaging devices, as well as 2,706 short videos, covering 1,016 subjects with diverse ages, genders and ethnicities. It encompasses 26 types of ICAO non-compliant scenarios, such as occlusions, abnormal lighting and other similar cases, and is accompanied by 900,000 manual annotations and 400,000 automatic annotations. The data is collected via multiple devices, with real-world scenarios simulated under controlled conditions, and has applications in fields including automated passport photo compliance detection and fairness enhancement of facial recognition systems.
DFIC数据集概述
数据集基本信息
- 数据集名称:DFIC (Towards a balanced facial image dataset for automatic ICAO compliance verification)
- 数据规模:包含约58,000张标注图像和2,706个视频,涉及超过1,000名对象。
- 核心内容:涵盖广泛的不合规条件以及合规肖像,旨在促进自动ICAO(国际民用航空组织)合规性验证方法的开发。
获取与使用条款
- 获取方式:需通过填写表单申请(https://forms.gle/pLfb6eRAmfbgNNVU9)。
- 使用限制:仅可用于非商业研究目的。
- 禁止行为:不得出于任何商业目的复制、销售、交易或利用任何部分图像及衍生数据;不得进一步复制、发布或分发数据集的任何部分(同一组织内部单站点使用允许复制)。
- 用户义务:同意数据集中的个人可根据GDPR行使权利,包括访问、更正、删除或限制处理其个人数据,以及反对处理的权利。
相关资源
- 预训练模型下载地址:https://drive.google.com/drive/folders/1l6OAiyM0k93KJQ7G6ORzoX4oryrifYUc?usp=sharing
- 许可证文件:https://github.com/visteam-isr-uc/DFIC/blob/main/LICENSE.md
- 数据集结构说明文件:https://github.com/visteam-isr-uc/DFIC/blob/main/data/DFIC/README.md
- 关联论文:https://arxiv.org/abs/2602.10985
引用格式
若使用本代码库、方法、模型或数据集,请引用:
@misc{gonçalves2026dficbalancedfacialimage, title={DFIC: Towards a balanced facial image dataset for automatic ICAO compliance verification}, author={Nuno Gonçalves and Diogo Nunes and Carla Guerra and João Marcos}, year={2026}, eprint={2602.10985}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2602.10985}, }




