hSDB-instrument: Instrument Localization Database for Laparoscopic and Robotic Surgeries

Automated surgical instrument localization is an important technology to understand the process and in order analyze them provide meaningful guidance during surgery or index after surgeon. We introduce a new dataset that reflects kinematic characteristics of instruments for automated videos. The hSDB(hutom Surgery DataBase)-instrument consists information from 24 cases laparoscopic cholecystecomy robotic gastrectomy. Localization all provided form bounding box object detection. To handle class imbalance problem between instruments, synthesized modeled Unity 3D models are included as training data. Besides, data, polygon annotation enable instance segmentation tool. reflect they annotated with head body parts head, wrist, separately. Annotation data assistive tools (specimen bag, needle, etc.) frequently used also included. Moreover, we statistical on hSDB-instrument baseline performances detection networks trained by MMDetection library resulting analyses.