ROCO

该数据集名为ROCO，包含了超过81,000对放射学图像与描述的配对，是预训练视觉-语言模型中广泛使用的医学数据集。每个图像与描述的配对还包括用于遮蔽语言建模的关键词和语义类型。该数据集的规模达到81,000多对图像与描述的配对，其任务包括图像-描述配对以及遮蔽语言建模。

This dataset is named ROCO. It contains over 81,000 pairs of radiological images and their corresponding textual descriptions, and is a widely adopted medical dataset for pre-trained vision-language models. Each image-description pair additionally provides keywords and semantic types for masked language modeling. The supported tasks of this dataset include image-text matching and masked language modeling.