OODFace
收藏资源简介:
OODFace数据集由北京航空航天大学人工智能研究所和中国信息通信研究院创建,旨在评估面部识别模型在常见损坏和外观变化下的鲁棒性。该数据集包含30个OOD场景,分为9个主要类别,涵盖了20个子类别的常见损坏和10个子类别的外观变化,每个子类别有5个严重级别,共计150个独特的损坏和变化场景。通过模拟这些挑战,数据集旨在全面评估现有面部识别模型在真实世界应用中的鲁棒性,并为未来提高模型鲁棒性提供指导。
The OODFace dataset was developed by the Institute of Artificial Intelligence at Beihang University and the China Academy of Information and Communications Technology, with the objective of evaluating the robustness of facial recognition models against common corruptions and appearance variations. This dataset includes 30 out-of-distribution (OOD) scenarios, which are categorized into 9 main classes, covering 20 subclasses of common corruptions and 10 subclasses of appearance variations. Each subclass has 5 severity levels, resulting in a total of 150 unique corruption and variation scenarios. By simulating these real-world challenges, the dataset aims to comprehensively evaluate the robustness of existing facial recognition models in practical applications, and provide guidance for future efforts to improve model robustness.

- 1OODFace: Benchmarking Robustness of Face Recognition under Common Corruptions and Appearance Variations北京航空航天大学人工智能研究所,中国信息通信研究院 · 2024年



