SurgiSR4K
收藏资源简介:
SurgiSR4K是一个原生4K分辨率(3840×2160像素)的内窥镜视频数据集,专为机器人辅助的微创手术设计。它包含了真实的机器人辅助腹腔镜手术场景,支持多种具有挑战性的条件,如镜面反射、工具遮挡、出血、烟雾、组织变形等。数据集支持超分辨率、工具检测、深度估计、分割等多种任务。
SurgiSR4K is a native 4K resolution (3840×2160 pixels) endoscopic video dataset designed specifically for robot-assisted minimally invasive surgery. It contains authentic robot-assisted laparoscopic surgery scenarios, and supports various challenging conditions such as specular reflection, tool occlusion, bleeding, smoke, tissue deformation, etc. The dataset supports multiple tasks including super-resolution, tool detection, depth estimation, segmentation, and more.
SurgiSR4K数据集概述
数据集基本信息
- 名称: SurgiSR4K
- 类型: 高分辨率内窥镜视频数据集
- 分辨率: 原生4K (3840×2160像素)
- 应用领域: 机器人辅助微创手术(MIS)的计算机视觉研究
- 许可证: CC BY-NC 4.0
- 论文: SurgiSR4K: A High-Resolution Endoscopic Video Dataset for Robotic-Assisted Minimally Invasive Procedures
数据集特点
- 高分辨率: 所有视频以3840×2160像素原生4K分辨率捕获
- 真实场景: 真实的机器人辅助腹腔镜手术场景
- 多样化挑战条件: 包含镜面反射、工具遮挡、出血、烟雾和组织变形等复杂情况
- 多任务支持: 支持超分辨率、器械检测、深度估计、分割等多种任务
数据集结构
SurgiSR4K/ ├── LICENSE ├── README.md ├── docs/ │ └── DATASET_ORGANIZATION.md ├── data/ │ ├── images/ │ │ ├── 3840x2160p/ # 4K分辨率帧(ground truth) │ │ ├── 960x540p/ # 中等分辨率帧 │ │ └── 480x270p/ # 低分辨率帧(输入) │ └── videos/ │ └── 3840x2160_30fps/ # 30FPS的4K源视频 ├── scripts/ │ ├── split.py └── ...
数据规格
- 分辨率级别: 480×270p, 960×540p, 3840×2160p
- 工具类别: 1tool, 2tool, 3tool, 4tool(复杂度指示器)
- 命名模式:
vid_{ID}_{resolution}_{tool}_{frame}.png - 总帧数: 2,400(25个视频中每个分辨率800帧)
主要任务
- 超分辨率(SR):
- 输入: 低分辨率帧(480p, 960p, 1080p)
- 目标: 原生4K分辨率帧
- 评估指标: PSNR, SSIM, LPIPS和感知质量指标
下游应用
- 实例分割
- 手术工具检测(带边界框)
- 深度估计
- 工具分割
- 3D重建
下载信息
引用格式
bibtex @article{jiang2025surgisr4k, author = {Fengyi Jiang and Xiaorui Zhang and Lingbo Jin and Ruixing Liang and Yuxin Chen and Adi Chola Venkatesh and Jason Culman and Tiantian Wu and Lirong Shao and Wenqing Sun and Cong Gao and Hallie McNamara and Jingpei Lu and Omid Mohareri}, title = {SurgiSR4K: A High‐Resolution Endoscopic Video Dataset for Robotic‐Assisted Minimally Invasive Procedures}, journal = {arXiv preprint arXiv:2507.00209}, year = {2025}, volume = {2507.00209}, doi = {10.48550/arXiv.2507.00209}, url = {https://arxiv.org/abs/2507.00209} }
联系方式
- 主要联系人: Fengyi Jiang (fengyi_jiang@alumni.brown.edu)
- 数据集咨询: ray.zhang@intusurg.com




