SPHINX
收藏资源简介:
SPHINX是一个用于多维度评估移动UI导航的基准数据集,由北京大学和腾讯公司等机构联合开发。该数据集包含284个任务,涵盖了112个流行移动应用的17个类别,旨在评估模型在目标理解、知识规划、基础操作和指令遵循等方面的能力。SPHINX通过自动化的基准测试套件,支持多种输入模态和操作空间,能够对模型进行全面的评估。数据集的应用领域主要集中在移动应用的自动化测试和UI导航,旨在解决现有模型在复杂UI导航任务中的不足,并为未来的研究提供方向。
SPHINX is a benchmark dataset for multidimensional evaluation of mobile UI navigation, jointly developed by Peking University, Tencent and other institutions. This dataset contains 284 tasks covering 112 popular mobile applications across 17 categories, aiming to evaluate models' capabilities in target understanding, knowledge planning, basic operations and instruction following. SPHINX supports multiple input modalities and operation spaces via an automated benchmark testing suite, enabling comprehensive evaluation of models. The application fields of this dataset mainly focus on automated testing of mobile applications and UI navigation, aiming to address the shortcomings of existing models in complex UI navigation tasks and provide directions for future research.




