CelebA-Gender
收藏资源简介:
CelebA-Gender数据集是基于CelebA人脸数据集构建的,专门用于评估联邦学习方法在数据分布差异较大的场景中的性能。该数据集通过确保不同面部属性的分布差异显著,同时保持男女性别类别平衡,模拟了高协变量偏移的情况。数据集的创建旨在解决联邦学习中由于客户端数据分布不一致导致的模型聚合不稳定问题,适用于性别分类等实际应用场景。
The CelebA-Gender dataset is constructed based on the CelebA face dataset, and is specifically designed to evaluate the performance of federated learning methods in scenarios with large data distribution discrepancies. It simulates high covariate shift scenarios by ensuring significant distribution differences across various facial attributes while maintaining a balanced gender category distribution. The dataset is created to address the unstable model aggregation problem caused by inconsistent client data distributions in federated learning, and is suitable for practical application scenarios such as gender classification.




