FaceCaption-15M

该数据集是一个大规模、多样化且高质量的人脸图像及其自然语言描述集合，旨在推动以人脸为中心的研究任务。该数据集基于LAION-Face数据集构建而成，原始数据集中包含了超过5000万张图像与文本的配对。在构建过程中，我们对图像进行了处理，分割出人脸区域，并采用自动标注方法结合大型语言模型生成了高质量的文本描述。该数据集包含了1500万对人脸图像与文本的配对，其研究任务主要针对人脸图像到文本的生成。

This dataset is a large-scale, diverse and high-quality collection of paired facial images and their natural language descriptions, designed to advance face-centric research tasks. Constructed upon the LAION-Face dataset, the original corpus contains more than 50 million image-text pairs. During the dataset construction, we processed the raw images to extract facial regions, and utilized automatic annotation methods combined with large language models to generate high-quality textual descriptions. This dataset ultimately includes 15 million facial image-text pairs, with its targeted research tasks primarily focusing on facial image-to-text generation.