EgoX jointly reconstructs hands, objects and camera from egocentric video, cutting hand error 65.6%
jiqizhixin · x · 2026-11-09
A preprint from Ropedia, NTU, Tsinghua and PKU introduces EgoX, which brings camera motion, depth, hands and objects into a shared metric world coordinate system and processes monocular egocentric video jointly, with a data engine supplying aligned supervision. On the accompanying EgoXBench it reports 65.6% lower hand reconstruction error and 42.9% lower depth error than the strongest baselines. Architecture uses DINOv3 ViT-L/16 with entity tokenization and Scene-Entity Blocks feeding depth, scale, camera, hand and object heads. Limitations: inference requires first-frame object masks, and reported gains apply to specific reconstruction metrics only.
More from Research
- New benchmarks DecoyCount and ScienceVision probe whether AI sees or reasons — liangchen5518 · 2026-11-09
- AI chip design demo: one prompt for a 64-bit RISC-V processor under 3W TDP — satnam6502 · 2026-11-09
- Video-guided machine translation paper lifts low-resource Indic languages by +1.75 BLEU — kalyan_kpl · 2026-11-09
- Peer-reviewed paper applies truth maintenance systems to LLM memory belief revision — letandrewcook · 2026-11-09
- A cathedral that sings: 1D acoustic sim shows voices fighting their own resonant body — Sauers_ · 2026-11-09
- Survey by 30+ authors maps modern NLP tokenization, from subword wins to its flaws — rdesh26 · 2026-11-09