3D-STARES: A Dataset for Surgical Tool Detection and Classification in Retinal Eye Surgeries

3D-STARES (3D Surgical Tool Annotation for Retinal Eye Surgeries) is a novel dataset designed to support the development of algorithms for surgical tool detection, classification, and depth estimation in retinal eye surgeries. The dataset contains intraoperative images captured from inside the eye during surgical procedures. It features comprehensive annotations for the specific type of surgical tool used, bounding boxes for the tool tip, and the approximate categorical distance of the tool tip to the retina. A key innovation of this dataset is the inclusion of the distance to the retinal surface alongside 2D horizontal location. This distinguishes it from previous datasets and enables the effective training of distance estimators. To ensure high-quality ground truth, the data underwent a multi-stage annotation process involving six human annotators, including three surgeons and three non-surgeons, utilizing a majority voting scheme to maximize accuracy. The authors have also provided preliminary results using state of the art object detection models for tool detection and depth classification to serve as a baseline for future studies. By offering these resources, 3D-STARES aims to facilitate research in computer-assisted interventions and surgical skill assessment, ultimately contributing to improved surgical training and patient outcomes.