BUPT-BalancedFace, BUPT-GlobalFace, SynGAN, CPD-25
收藏资源简介:
本研究涉及四个数据集:BUPT-BalancedFace和BUPT-GlobalFace为真实数据集,分别包含1.3百万和2百万张图像,旨在研究面部识别模型的偏见;SynGAN和CPD-25为合成数据集,分别包含50万和5万张图像,用于训练面部识别系统。这些数据集通过大规模属性分类器(MAC)进行注释,以分析真实与合成数据集之间的差异。数据集的应用领域主要集中在面部识别技术的改进和偏见研究,旨在通过合成数据减少对真实数据的依赖,提高模型的性能和公平性。
This study involves four datasets: BUPT-BalancedFace and BUPT-GlobalFace are real-world datasets, containing 1.3 million and 2 million images respectively, and are designed to study the bias in facial recognition models. SynGAN and CPD-25 are synthetic datasets, containing 500,000 and 50,000 images respectively, and are used for training facial recognition systems. These datasets are annotated using the Massive Attribute Classifier (MAC) to analyze the differences between real-world and synthetic datasets. The application fields of these datasets mainly focus on the improvement of facial recognition technology and bias research, aiming to reduce the reliance on real-world data through synthetic data, thereby enhancing model performance and fairness.




