Casual Conversations

该数据集包含了大约15段每位受试者一分钟的录像，共涉及3,011名受试者，这些录像在不同美国城市录制而成。数据集对年龄、性别、肤色以及低光照环境等进行了标注。此外，数据集中还包含了受试者自我识别的年龄和性别标注、Fitzpatrick肤色标注，并且在性别和年龄组之间分布均衡。该数据集由3,011名受试者和超过45,000段视频组成，其任务是评估计算机视觉和音频模型在不同人口统计学特征下的公平性和鲁棒性。

This dataset includes approximately 15 one-minute video clips per subject, with a total of 3,011 participants, and all footage was recorded across multiple U.S. cities. The dataset provides annotations for attributes such as age, gender, skin tone, and low-light recording environments. Additionally, it contains self-identified age and gender annotations from the subjects, as well as Fitzpatrick skin type annotations, and maintains a well-balanced distribution across gender and age groups. Comprising 3,011 subjects and over 45,000 video clips in total, this dataset is designed to evaluate the fairness and robustness of computer vision and audio models across diverse demographic characteristics.