agent-studio/CriticBench

CriticBench是一个用于评估虚拟代理成功检测能力的AgentStudio数据集。该数据集包含指令、动作序列等信息，动作序列中包括动作ID、操作前后的图像、操作类型、边界框等元数据。数据集被拆分为训练集，共有350个示例，大小为1,638,036,122字节。数据集适用于视觉问题回答任务，使用MIT许可。

CriticBench is a dataset in AgentStudio for evaluating the success detection abilities of virtual agents. The dataset includes instructions, action sequences, which consist of action ID, images before and after the action, type of operation, bounding boxes, and other metadata. The dataset is split into a training set with 350 examples, totaling 1,638,036,122 bytes in size. It is suitable for visual question answering tasks and is licensed under MIT.