Ref-EndoVis17 和 Ref-EndoVis18
收藏资源简介:
Ref-EndoVis17 和 Ref-EndoVis18 数据集是在 EndoVis17 和 EndoVis18 基础上构建的,用于计算机辅助手术中的场景分割。这些数据集包含了丰富的场景分割标注,包括手术器械和组织的标注,旨在提高手术质量并优化患者护理结果。数据集由新加坡国立大学的研究团队创建,并通过引用分割技术实现,允许外科医生通过文本表达式交互式地识别和跟踪特定的对象。ReSurgSAM2 方法在 Ref-EndoVis17 和 Ref-EndoVis18 数据集上的实验表明,与现有方法相比,该数据集能够提供更高的准确性和效率,并能够在实时情况下以 61.2 FPS 运行。
The Ref-EndoVis17 and Ref-EndoVis18 datasets are developed based on EndoVis17 and EndoVis18, targeting scene segmentation tasks in computer-assisted surgery. These datasets feature rich scene segmentation annotations covering surgical instruments and tissues, with the goal of enhancing surgical quality and optimizing patient care outcomes. Created by a research team from the National University of Singapore, the datasets employ reference segmentation technology, allowing surgeons to interactively identify and track specific objects via text expressions. Experiments conducted on the Ref-EndoVis17 and Ref-EndoVis18 datasets using the ReSurgSAM2 method show that, compared with existing state-of-the-art approaches, this dataset suite delivers higher accuracy and efficiency, and can run at 61.2 FPS in real-time scenarios.
ReSurgSAM2 数据集概述
基本信息
- 数据集名称: ReSurgSAM2
- 相关论文: ReSurgSAM2: Referring Segment Anything in Surgical Video via Credible Long-term Tracking
- 作者: Haofeng Liu, Mingqi Gao, Xuxiao Luo, Ziyue Wang, Guanyi Qin, Junde Wu, Yueming Jin
- 会议: Early accepted by MICCAI 2025
数据集特点
- 应用领域: 计算机辅助手术中的手术场景分割
- 框架特点:
- 两阶段手术参考分割框架
- 利用SAM2进行文本参考目标检测
- 采用Cross-modal Spatial-Temporal Mamba (CSTMamba)进行精确检测和分割
- 包含Credible Initial Frame Selection (CIFS)策略
- 采用Diversity-driven Long-term Memory (DLM)保持可信和多样化的记忆库
- 实时运行速度: 61.2 FPS
数据来源
- 基础数据集:
使用说明
- 预处理步骤: 参见datasets/README.md
致谢
- 基于segment anything 2框架
- 使用CLIP实现




