Never Look Back: Understanding Persistence in 3D Object Memory from Egocentric Videos
Ledger keeps 3D object records alive after objects leave the camera view.
The paper tests an embodied-assistant memory built from egocentric video, storing object locations, histories, and short contextual descriptions. It clusters observations by resting place and only records moves after repeated evidence, aiming to avoid noise. Reported gains include HD-EPIC accuracy rising from 29.7% to 42.6% and UCS-Bench from 33.8% to 38.5%. The authors also find failures when streams span multiple scenes, with per-scene construction recovering some lost performance. ArXiv · AI/CL/LG's note
The paper tests an embodied-assistant memory built from egocentric video, storing object locations, histories, and short contextual descriptions. It clusters observations by resting place and only records moves after repeated evidence, aiming to avoid noise. Reported gains include HD-EPIC accuracy rising from 29.7% to 42.6% and UCS-Bench from 33.8% to 38.5%. The authors also find failures when streams span multiple scenes, with per-scene construction recovering some lost performance. ArXiv · AI/CL/LG's note
score 5