Surgical Dataset Generation Based on 3D Gaussian Splatting
收藏资源简介:
本研究介绍了一种基于3D高斯喷溅技术生成手术图像数据集的新方法,由利兹大学开发。该数据集包含1568条图像数据,通过精确编辑3D高斯模型,实现了手术场景和器械模型的独立训练与灵活组合。数据集的创建过程涉及从手术场景和器械中提取高斯表示,进行必要的编辑和融合,以生成高质量的合成手术场景。此数据集主要应用于神经网络训练,旨在提高机器人辅助微创手术中的自动化能力,特别是在手术器械的跟踪、检测和定位方面。
This study introduces a novel method for generating surgical image datasets based on 3D Gaussian Splatting technology, developed by the University of Leeds. This dataset comprises 1568 image samples, enabling independent training and flexible combination of surgical scene and instrument models through precise editing of 3D Gaussian models. The dataset creation process entails extracting Gaussian representations from surgical scenes and instruments, followed by requisite editing and fusion to produce high-quality synthetic surgical scenes. This dataset is primarily utilized for neural network training, with the goal of improving automation capabilities in robot-assisted minimally invasive surgery, especially regarding the tracking, detection and localization of surgical instruments.

- 1Realistic Surgical Image Dataset Generation Based On 3D Gaussian Splatting利兹大学 · 2024年



