平衡人脸数据集

平衡人脸数据集是由非洲数学科学研究所的Kidist Amde Mekonnen创建，旨在通过StyleGAN模型生成一个包含180,000张合成人脸图像的大型数据集，以解决现有数据集在不同人口统计群体中的不平衡问题。该数据集通过控制StyleGAN的生成过程，确保了各群体的均匀分布，并针对多种下游任务进行了标注，如眼睛状态、微笑和人口统计分类。创建过程涉及使用UTKFace数据集训练的分类器来指导图像生成，确保多样性和代表性。此数据集的应用领域广泛，主要用于提高面部识别系统的公平性和准确性，以及支持情感识别和面部表情分析等计算机视觉任务。

The Balanced Face Dataset was developed by Kidist Amde Mekonnen from the African Institute for Mathematical Sciences. It aims to generate a large-scale dataset containing 180,000 synthetic face images via the StyleGAN model to address the imbalance issue of existing datasets across diverse demographic groups. This dataset controls the StyleGAN generation process to ensure uniform distribution across all demographic groups, and is annotated for multiple downstream tasks including eye state, smiling status, and demographic classification. The creation process involved using a classifier trained on the UTKFace dataset to guide image generation, ensuring the dataset's diversity and representativeness. This dataset has broad application scenarios, primarily used to improve the fairness and accuracy of facial recognition systems, as well as support computer vision tasks such as emotion recognition and facial expression analysis.