ControlFace10k
收藏资源简介:
ControlFace10k是由南卫理公会大学智能系统与偏见审查实验室创建的一个用于人脸识别评估的开源数据集。该数据集包含10,008张人脸图像,涵盖3,336个独特的合成身份,这些身份在种族、性别和年龄上进行了平衡。数据集通过合成身份生成管道(SIG)创建,该管道能够生成高质量的合成身份图像,具有可控的姿态、面部特征和人口统计属性。ControlFace10k旨在解决现有数据集在种族、性别和年龄上的不平衡问题,为评估人脸识别算法在不同人口统计群体中的表现提供了一个平衡和全面的工具。
ControlFace10k is an open-source dataset for face recognition evaluation, developed by the Smart Systems and Bias Review Lab at Southern Methodist University. This dataset comprises 10,008 facial images covering 3,336 unique synthetic identities that are balanced across race, gender, and age groups. It is constructed via the Synthetic Identity Generation (SIG) pipeline, which can generate high-quality synthetic facial images with controllable poses, facial features, and demographic attributes. ControlFace10k aims to address the racial, gender, and age imbalance issues present in existing datasets, providing a balanced and comprehensive benchmark for evaluating the performance of face recognition algorithms across diverse demographic groups.




