dad3131/GUIOdyssey

GUIOdyssey是一个用于训练和评估跨应用导航代理的综合数据集。该数据集包含来自6种移动设备的8,334个任务片段，涵盖6种跨应用任务类型、212个应用和1.3K个应用组合。数据集提供了丰富的语义注释，包括每个步骤的细粒度低级指令、图像描述、动作意图和上下文回顾。此外，还为每个点击的元素提供了边界框，以实现更准确的评估。数据集支持多种分割方式（随机分割、任务分割、设备分割和应用分割），以评估代理在不同场景下的泛化能力。

GUIOdyssey is a comprehensive dataset for training and evaluating cross-app navigation agents. It consists of 8,334 episodes from 6 mobile devices, spanning 6 types of cross-app tasks, 212 apps, and 1.3K app combos. The dataset features rich semantic annotations, including fine-grained low-level instructions, image descriptions, action intentions, and context review for each step. Additionally, it provides bounding boxes for each clicked element, enabling more accurate evaluation. The dataset supports multiple splitting methods (random_split, task_split, device_split, and app_split) to evaluate the agents generalization capabilities across different scenarios.