LostFound

[RA-L] Lost & Found dynamically tracks object poses from egocentric videos while updating a scene graph, enabling richer semantic 3D understanding for robotic downstream tasks.