DataVisualWorkflow

DataVisualWorkflow 是由摩根大通人工智能研究团队创建的一个用于计算机用户界面理解的新数据集。该数据集包含14345帧视频，记录了用户在计算机上执行的一系列操作，每帧显示桌面内容。数据集的创建旨在推动计算机UI理解的研究，特别是针对复杂的计算机UI环境，这些环境中的界面可能混乱且变化多样。DataVisualWorkflow 数据集不仅用于半监督学习，还可能应用于无监督学习、动作识别、学习演示、开放集学习和异常行为检测等领域。

DataVisualWorkflow is a novel dataset for computer user interface understanding created by the Artificial Intelligence Research Team of JPMorgan Chase. This dataset contains 14,345 video frames that record a series of operations performed by users on computers, with each frame displaying desktop content. The dataset is developed to advance research on computer UI understanding, particularly for complex computer UI environments where interfaces may be chaotic and highly variable. The DataVisualWorkflow dataset can be used not only for semi-supervised learning, but also for potential applications including unsupervised learning, action recognition, learning from demonstration, open-set learning, and anomalous behavior detection.