lodestones/pixelprose

PixelProse是一个包含超过1600万条由先进视觉语言模型生成的密集图像描述的综合数据集。数据来源于CommonPool、CC12M和RedCaps，涵盖了详细的图像描述。数据集以Parquet文件格式提供，包含图像URL、描述、毒性评分等多个字段。

PixelProse is a comprehensive dataset containing over 16 million dense image captions generated by state-of-the-art vision-language models. It is sourced from CommonPool, CC12M and RedCaps, and includes detailed image captions. The dataset is provided in Parquet file format, with multiple fields including image URLs, captions, toxicity scores and more.