RoCOCO

RoCOCO数据集是由首尔国立大学电气与计算机工程学院的研究人员创建的，旨在通过对抗性文本和图像来评估图像文本匹配模型的鲁棒性。数据集包含50,000条数据，其中一半是原始的MS-COCO数据，另一半是通过在文本中替换单词或在图像中混合不同图像生成的对抗性数据。该数据集的应用领域主要集中在提高视觉语言模型的鲁棒性，以及开发更多样化的跨模态检索任务的压力测试方法。

The RoCOCO dataset was constructed by researchers from the School of Electrical and Computer Engineering, Seoul National University, to evaluate the robustness of image-text matching models using adversarial texts and images. It comprises 50,000 samples in total, with half being original MS-COCO data, and the remaining half consisting of adversarial samples generated by replacing words in text descriptions or blending different images into the original ones. The primary application scenarios of this dataset focus on enhancing the robustness of vision-language models and developing more diverse stress test methods for cross-modal retrieval tasks.