Casual Conversations Dataset

Casual Conversations Dataset是由Facebook AI创建的一个视频数据集，旨在评估AI模型在不同年龄、性别、肤色和光照条件下的公平性和鲁棒性。该数据集包含来自美国多个州的3,011名不同年龄、性别和肤色的成年人录制的超过45,000个视频，平均每人15个视频。数据集中的年龄和性别标签由参与者自行提供，肤色标签则由训练有素的注释者使用Fitzpatrick肤色类型量表标注。此外，还提供了低光照条件下的视频标注。该数据集特别适用于评估深度伪造检测和其他面部属性分类模型的性能，尤其是在处理不同背景人群时的公平性问题。

The Casual Conversations Dataset is a video dataset created by Facebook AI, designed to evaluate the fairness and robustness of AI models across varying ages, genders, skin tones and lighting conditions. It contains over 45,000 video recordings of 3,011 adults from multiple U.S. states, with diverse ages, genders and skin tones, averaging 15 videos per participant. Age and gender labels are self-reported by participants, while skin tone labels are annotated by trained annotators using the Fitzpatrick Skin Type Scale. Additionally, video annotations under low-light conditions are provided. This dataset is particularly suitable for evaluating the performance of deepfake detection and other facial attribute classification models, especially regarding fairness issues when processing populations with varied backgrounds.