pengyueli/RAF_FairFace

该数据集是RAF-Ready FairFace，一个经过处理的版本，专为论文《Retrieval-Augmented Dataset Assembly for Fair Clustering》中的RAF框架设计。RAF是一个检索增强的数据集组装框架，用于下游公平聚类任务，旨在通过从外部数据源检索和选择样本来减少少数群体的代表性不足，从而在聚类前改善数据分布。数据集基于原始FairFace数据集，后者是一个用于平衡种族、性别和年龄分析的人脸属性数据集。处理后的版本包括查询数据集（作为初始数据集D_q）、外部源文件、源元数据、视觉嵌入和敏感群体标签，这些文件支持在视觉数据域中运行RAF实验，以评估在嵌入空间中多数-少数群体分布对齐的改进。This dataset is RAF-Ready FairFace, a processed version designed for the RAF framework in the paper Retrieval-Augmented Dataset Assembly for Fair Clustering. RAF is a retrieval-augmented dataset assembly framework for downstream fair clustering, aiming to reduce minority under-representation by retrieving and selecting real external samples to improve data distribution before clustering. It is based on the original FairFace dataset, a face attribute dataset for balanced race, gender, and age analysis. The processed version includes the query dataset (used as the initial dataset D_q), external source files, source metadata, visual embeddings, and sensitive-group labels, which support running RAF experiments in the visual-data domain to evaluate improvements in majority-minority distributional alignment in the embedding space.