REVO-LION

REVO-LION是一个综合性的视觉-语言指令调优数据集，由上海交通大学等机构的研究人员构建。该数据集通过从多个现有数据集中精选出高质量样本组成，旨在支持开发全能型视觉-语言模型。REVO-LION不仅包含用于模型训练的样本，还提供了一个评估集，用于作为未来研究的便捷基准。该数据集的设计强调了样本质量的重要性，通过模型无关和人工无关的评估方法确保数据集的质量，适用于解决视觉-语言理解领域的复杂问题。

REVO-LION is a comprehensive vision-language instruction tuning dataset constructed by researchers from institutions such as Shanghai Jiao Tong University. Composed of high-quality samples curated from multiple existing datasets, this dataset is designed to support the development of universal vision-language models. REVO-LION not only includes training samples for model development but also provides an evaluation set that can serve as a convenient benchmark for future research. The design of this dataset emphasizes the significance of sample quality, ensuring dataset reliability through model-agnostic and human-independent evaluation approaches, making it suitable for addressing complex problems in the field of vision-language understanding.