POV-Surgery
收藏资源简介:
POV-Surgery是由苏黎世联邦理工学院创建的一个大规模合成数据集,专注于从第一人称视角进行手术活动中的手和工具姿态估计。该数据集包含53个序列,共计88,329帧,涵盖高分辨率RGB-D视频流,具有活动标注、精确的3D和2D手-物体姿态标注以及2D手-物体分割掩码。数据集通过创新的合成数据生成管道创建,模拟了真实手术场景中的手部动作和工具使用,包括不同手术手套和三种骨科手术工具。该数据集旨在解决现有数据集在手术环境中手和工具姿态估计的不足,特别是在处理血迹手套和反射金属工具时的挑战。
POV-Surgery is a large-scale synthetic dataset developed by ETH Zurich, focusing on first-person view hand and tool pose estimation during surgical activities. It contains 53 sequences totaling 88,329 frames, covering high-resolution RGB-D video streams, with activity annotations, precise 3D and 2D hand-object pose annotations, as well as 2D hand-object segmentation masks. The dataset is built via an innovative synthetic data generation pipeline that simulates hand movements and tool utilization in realistic surgical scenarios, including various surgical gloves and three orthopedic surgical tools. This dataset aims to address the limitations of existing datasets for hand and tool pose estimation in surgical environments, particularly the challenges posed by blood-stained gloves and reflective metal surgical tools.




