RoCOCO

该数据集旨在为图像-文本匹配模型建立一个健壮性基准，通过将对抗性文本和图像插入搜索池中来评估这些模型的脆弱性。此外，该数据集还包含了由修改原始COCO图像标题生成的欺骗性图像和标题，以评估ITM模型的鲁棒性。该数据集的任务是针对图像-文本匹配（ITM）。

This dataset is designed to establish a robustness benchmark for image-text matching (ITM) models, where adversarial texts and images are inserted into the search pool to evaluate the vulnerability of such models. Additionally, this dataset includes deceptive images and captions generated by modifying the original COCO image captions, aiming to assess the robustness of ITM models. The core task targeted by this dataset is image-text matching (ITM).