agent-studio/IDM-Single

IDM-Single是IDMBench的一部分，用于评估智能体在没有动作标签的互联网规模视频中的学习能力。该数据集特别要求视觉语言模型回答两个屏幕截图之间的动作转换。这是一个测试集，不应用于训练。

IDM-Single is part of the IDMBench for evaluating agents ability to learn from Internet-scale videos without action labels. Specifically, this dataset requires Visual Language Models to answer the action transition between two screenshots. This is a test set and should not be used for training.