NLVR2 and OKVQA-S

NLVR2 is a challenging VQA dataset that requires the model to compare, locate, and count objects based on the given question and images. OKVQA-S is a challenging category of OKVQA that requires the model to use external knowledge, such as common sense, and facts, to answer questions about different topics and scenarios.

NLVR2是一个极具挑战性的视觉问答（Visual Question Answering, VQA）数据集，要求模型基于给定的问题与图像完成对象的比较、定位与计数任务。OKVQA-S是OKVQA旗下一类极具挑战性的任务子集，其要求模型借助常识、事实等外部知识，对不同主题与场景下的问题进行作答。