OPPOer/AndesUI_Benchmark

AndesUI_Benchmark数据集是一个用于用户界面理解的基准数据集，包含训练集和测试集。数据集涵盖三种主要任务类型：定位任务、指示任务和问答任务。定位任务是根据控件描述预测边界框坐标；指示任务是根据边界框预测相应的控件描述；问答任务是回答自然语言问题，通过预测需要点击的边界框坐标。数据集子集以JSON和TSV格式提供，包含相对路径的图像路径。

The AndesUI_Benchmark dataset is a benchmark for user interface understanding, consisting of a training set and a test set. It covers three main task types: Grounding Task, Referring Task, and QA Task. The Grounding Task predicts bounding box coordinates from a widget description, the Referring Task predicts the corresponding widget description from a bounding box, and the QA Task answers natural language questions by predicting the bbox coordinates to be clicked. The subsets of the dataset are provided in both JSON and TSV formats with relative paths for image locations.