ZhuOnR/guiact_smartphone_test

GUIAct智能手机数据集是一个用于智能手机GUI导航任务的多步骤动作指令集合。该数据集是通过转换和适配AITW（Androids in the Wild）数据集的一个子集创建的，特别关注带有“General”标签的数据。数据集包含与智能手机截图配对的序列动作（如点击、滑动、输入），旨在训练视觉语言模型以导航智能手机界面。每个指令需要多个动作来完成，动作映射到包括位置信息的标准化格式。该数据集是更大的GUICourse集合的关键组成部分，用于训练多功能GUI代理。数据集包含9,157个多步骤动作指令，对应约67,000个训练样本。每个样本包括智能手机截图、一个或多个要在该截图上执行的动作以及描述要完成的任务的自然语言指令。动作空间包括标准化的动作，如点击、滑动、输入、进入和回答。

The GUIAct Smartphone dataset is a collection of multi-step action instructions for smartphone GUI navigation tasks. It was created by converting and adapting a subset of the AITW (Androids in the Wild) dataset, specifically focusing on data with the General tag. The dataset contains sequences of actions (such as tap, swipe, input) paired with smartphone screenshots, designed to train vision language models to navigate smartphone interfaces. Each instruction requires multiple actions to complete the task, with actions mapped to a standardized format that includes position information. The dataset serves as a crucial component of the larger GUICourse collection for training versatile GUI agents. The dataset contains 9,157 multi-step action instructions which correspond to approximately 67,000 training samples. Each sample consists of a smartphone screenshot, one or more actions to be performed on that screenshot, and a natural language instruction describing the task to be accomplished. The action space includes standardized actions such as tap, swipe, input, enter, and answer.