AI
InfiniHand: Streaming World-Space Hand Motion Estimation from Egocentric Video
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 15:03:06 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 07:18:25 UTC.
Briefings
InfiniHand is an end-to-end streaming framework for world-space hand motion estimation that jointly recovers 3D hand geometry, camera trajectories, and hand locations from uncalibrated egocentric video. It replaces cascading pose estimators with a unified architecture integrating persistent spatiotemporal memory and hand-centered visual features. The model was trained using approximately 5,000 hours of egocentric data across multiple public datasets, achieving a 21.4% reduction in ARCTIC PA-p error compared to ViDiHand and significantly mitigating world-space drift. Performance on in-the-wild videos reaches 11.19 FPS, exceeding HaWoR throughput by more than double.
Key points
- InfiniHand is an end-to-end streaming framework that jointly estimates MANO hand parameters, camera trajectories, and hand locations directly from uncalibrated egocentric video using persistent spatiotemporal memory.
- The model aggregates approximately 5,000 hours of egocentric data across multiple public datasets to train robust camera-space hand priors before extending to streaming world-space reconstruction.
Context and limitations
Evaluation results are limited to in-domain benchmarks and in-the-wild generalization; specific quantitative metrics for world-space drift reduction are not fully quantified beyond the stated 21.4% ARCTIC PA-p reduction.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 15:03:06 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifhiy5vtzrkdieyuwv6ersiqq7ql3vdj3w2ff2d7wslt365afpahq