This repository contains example code associated with our paper.
At present, only the inference code and sample datasets are available. Comprehensive training details, codes, and resources will be made available upon the paper's acceptance.
We present a novel geometric approach with a streamlined yet comprehensive representation of both surgical targets and instruments. It provides a more robust interaction representation with high-level semantics. We introduce an adaptive graph learning to obtain a dynamic understanding of surgical procedures. Additionally, we propose an adaptive multi-horizon learning strategy to offer global insights for each horizon and balance learning objectives across various time spans.
python main.py
For any queries, please feel free to contact Xiatian Zhang at Xiatian.Zhang@durham.ac.uk