FairFace

FairFace是由加州大学洛杉矶分校创建的一个大型人脸图像数据集，包含108,501张图像，特别强调种族平衡。该数据集定义了7个种族群体：白人、黑人、印度人、东亚人、东南亚人、中东人和拉丁美洲人。图像主要从YFCC-100M Flickr数据集中收集，并由人工标注种族、性别和年龄。FairFace旨在解决现有公开人脸数据集在种族代表性上的不平衡问题，通过提供一个种族、性别和年龄都相对平衡的数据集，以提高人脸分析系统的准确性和公平性。

FairFace is a large-scale facial image dataset developed by the University of California, Los Angeles. It contains 108,501 images with a special emphasis on racial balance. The dataset defines seven racial groups: White, Black, Indian, East Asian, Southeast Asian, Middle Eastern, and Latino. Images are primarily collected from the YFCC-100M Flickr dataset and manually annotated with race, gender, and age. FairFace aims to address the racial representation imbalance issue in existing publicly available facial datasets. By providing a dataset with relatively balanced distributions across race, gender, and age, it seeks to improve the accuracy and fairness of facial analysis systems.

搜集汇总

背景与挑战

背景概述

FairFace是一个面部属性数据集，专注于提供平衡的种族、性别和年龄分布，旨在用于计算机视觉中的偏见测量和缓解研究。数据集包含经过裁剪和对齐的面部图像，并附带训练和验证标签，采用CC BY 4.0许可证开放使用。