Synthetic Counterfactual Faces Dataset

该数据集包含了一系列通过特定流程生成的合成人脸图像，该流程能够创造出不同的人种群体中，通过操纵各种面部特征所产生的反事实图像。所有图像都经过验证，确保它们代表同一身份，并经过筛选过程以去除扭曲的图像，从而支持计算机视觉模型在鲁棒性和公平性方面的研究。该数据集的规模为从4800个原始人脸生成的91,200个变换后的人脸，覆盖了8个不同的人种群体（4种族裔×2种性别）。其任务是用这些变换后的合成人脸来评估人脸识别系统，并通过这些图像理解图像生成模型。

This dataset comprises a set of synthetic facial images generated via a dedicated pipeline, which can produce counterfactual facial images by manipulating various facial attributes across different racial groups. All images have been validated to ensure they represent the same individual identity, and a filtering process has been applied to remove distorted samples, thereby supporting research on the robustness and fairness of computer vision models. The dataset consists of 91,200 transformed facial images generated from 4,800 original human faces, covering 8 distinct racial groups (4 ethnic groups × 2 genders). Its intended use is to evaluate facial recognition systems and facilitate the analysis of image generation models using these synthetic transformed faces.