Semantic Instance Extraction: CLIP and DINO Features for 3D Mapping

Details the O3D-SIM pipeline for VLN. It extracts open-set semantic instance information (masks, CLIP/DINO features) from RGB-D images