LEDGER: Persistent 3D Object Memory From Egocentric Video Lets You Ask Questions Without Replaying
mangahomanga · x · 2026-10-08
A Johns Hopkins team (with Suchi Saria and Rama Chellappa) introduces LEDGER, from the paper Never Look Back. The system builds a persistent 3D object memory from egocentric video: it remembers where objects were, when they moved, and what they held. After seeing a scene once, it can answer later questions directly from memory — no video replay needed. Project page and paper are linked in the original thread.
Related event: LEDGER Builds Persistent 3D Object Memory from Egocentric Video(2 posts)→
More from Research
- Psych researchers predict AI will break peer review and reshape social science publishing — steverathje2 · 2026-10-08
- Yaroslav Bulatov Traces the Origins of the Memory Wall in New Research Writeup — yaroslavvb · 2026-10-08
- Scale AI partners on HSS, a new benchmark for 'seeing what's in front of them' — AndrewDai · 2026-10-08
- Xiaomi's open-sourced MiMo RL environments leak answers in Git history for 2/3 of coding tasks — teortaxesTex · 2026-10-08
- BeFOND: encoder-free sparse coding matches Gemma Scope with 1000x fewer training tokens — gottapatchemall · 2026-10-08
- Rust+Vulkan Inference Backend Goes Cross-Vendor: Pick Kernels by CPU ISA, Not GPU Vendor — PhysicsDisastrous462 · 2026-10-08