CLiMB

CLiMB是一个由南加州大学开发的多模态持续学习基准数据集，包含四个视觉-语言任务、五个语言任务和四个视觉任务，总计13个任务。该数据集旨在研究多模态任务在持续学习设置下的挑战，并系统评估上游持续学习如何快速泛化到新的多模态和单模态任务。数据集通过GitHub提供，支持研究者探索新的持续学习算法和模型。

CLiMB is a multimodal continual learning benchmark dataset developed by the University of Southern California. It contains a total of 13 tasks, including four vision-language tasks, five language tasks and four vision tasks. This dataset aims to investigate the challenges of multimodal tasks under the continual learning setting, and systematically evaluate how upstream continual learning can rapidly generalize to new multimodal and unimodal tasks. The dataset is available via GitHub, enabling researchers to explore novel continual learning algorithms and models.