kaupane/human-recaption

该数据集包含240,146张以人类为主题的重新标注的图像，源自HumanCaption-HQ-311K数据集。它提供了高质量的双语（英文和中文）标注、美学评分和其他元数据，这些数据是通过Qwen2-VL模型生成的。每个条目包括原始图像URL、详细的双语标注（中文和英文）、美学质量评分、图像模式（RGB/灰度）和水印检测状态。标注内容特别关注人类属性，包括外貌、种族、服装、面部特征、姿势、表情和动作。This dataset contains 240,146 recaptioned images focusing on human subjects, derived from the HumanCaption-HQ-311K dataset. It provides high-quality bilingual (English and Chinese) captions, aesthetic scores, and other metadata generated using the Qwen2-VL model. Each entry includes the original image URL, detailed bilingual captions (Chinese and English), an aesthetic quality score, mode (RGB/Grayscale), and watermark detection status. The captions were generated to specifically focus on human attributes, including appearance, ethnicity, clothing, facial features, pose, expression, and action.