InstanceSeg4MIS

This is the code for surgical instrument instance segmentation by Jointly Optimizing Spatial Embeddings and Clustering Bandwidth.

Included models: ERFNet; OR-Unet; ViT (pre-trained by DINOV2) + MLA segmentation head

Training

Train on OR-Unet (RobustMIS 2019)

ROBOMIS_DIR=.../data/robo python src/train_unet.py

Testing

Test on ViT (pre-trained by DINOV2) + MLA segmentation head (RobustMIS 2019)

ROBOMIS_DIR=.../data/robo python src/test_mla.py

The pre-trained weights for this model can be downloaded here: Google Drive