MV-dVRK
收藏资源简介:
MV-dVRK是由马克斯·普朗克智能系统研究所等机构联合创建的首个多视角离体手术数据集,专为稀疏多视角三维重建评估而设计。该数据集包含静态和动态两个子集:静态子集涵盖8个场景、840帧高分辨率图像,通过密集运动恢复结构获取精确的表面几何与相机位姿,并经过工业扫描仪验证;动态子集则包含10个同步序列,记录复杂手术操作中的组织变形与器械运动。数据集构建依托扩展的达芬奇研究平台,实现了三路立体内窥镜的曝光同步采集,为评估单目、立体及多视角三维重建方法提供了标准基准。其应用聚焦于解决机器人辅助微创手术中空间感知的几何覆盖与位姿估计难题,推动稀疏输入下的高精度三维重建研究。
MV-dVRK is the first multi-view ex-vivo surgical dataset co-developed by the Max Planck Institute for Intelligent Systems and other institutions, designed specifically for the evaluation of sparse multi-view 3D reconstruction. The dataset contains two subsets: static and dynamic. The static subset covers 8 scenes and 840 high-resolution image frames, from which accurate surface geometry and camera poses are acquired via dense structure from motion (SfM), and validated by industrial scanners. The dynamic subset includes 10 synchronized sequences that record tissue deformation and instrument motion during complex surgical operations. Built upon an extended da Vinci research platform, the dataset enables synchronized exposure acquisition with three stereo endoscopes, serving as a standard benchmark for evaluating monocular, stereo, and multi-view 3D reconstruction methods. Its applications focus on addressing the challenges of geometric coverage and pose estimation for spatial perception in robot-assisted minimally invasive surgery, advancing high-precision 3D reconstruction research under sparse input conditions.

- 1MV-dVRK: A Multi-Viewpoint Benchmark for Spatial Surgical Perception马克斯·普朗克智能系统研究所; 苏黎世联邦理工学院; 爱尔博集团; 图宾根大学医院; 约翰霍普金斯大学 · 2026年




