In this work, we tackled the challenge of training models that can keep learning new surgical instruments over time, without forgetting the old ones. Data imbalance made this especially tricky, so we proposed a plug-and-play framework that balances the data using inpainting and blending techniques, and introduced a new loss function to reduce confusion between similar-looking tools.
Big thanks to our amazing team (Shifang Zhao, Long Bai, Kun Yuan, Feng Li, Jieming YU, Wenzhen Dong, Guankun Wang, Prof. Mobarak I. Hoque, Prof. Nicolas Padoy, Prof. Nassir Navab, Prof. Hongliang Ren) from CUHK, TUM, Strasbourg, and UCL. This collaboration truly brought together ideas from different corners of the world ๐
The paper is now online: https://lnkd.in/gemZFNUK Code coming soon ๐จโ๐ป
