Surg-FTDA
收藏资源简介:
Surg-FTDA数据集由斯特拉斯堡大学和慕尼黑工业大学的研究团队创建,旨在支持手术工作流分析任务。该数据集通过少量配对的图像-标签数据,结合文本驱动的方法,减少对大规模标注数据的依赖。数据集的内容包括手术场景的图像和对应的文本标签,用于训练和评估多模态基础模型。数据集的创建过程涉及少量数据锚点的选择和模态对齐,以缩小视觉和文本嵌入之间的差距。该数据集的应用领域主要集中在手术工作流分析,旨在提高手术效率和安全性,减少对专家标注的依赖。
The Surg-FTDA dataset was developed by research teams from the University of Strasbourg and the Technical University of Munich, aiming to support surgical workflow analysis tasks. By leveraging a small number of paired image-label samples and text-driven approaches, this dataset reduces the reliance on large-scale annotated datasets. The dataset comprises images of surgical scenes and their corresponding text labels, which are employed for training and evaluating multimodal foundation models. The development process of the dataset entails the selection of a small number of data anchors and modal alignment, to narrow the disparity between visual and text embeddings. The primary application domain of this dataset is surgical workflow analysis, with the objectives of enhancing surgical efficiency and safety, as well as reducing the dependence on expert annotations.




