Perceptron's egocentric video labeling system boosts hand reconstruction accuracy 12-25% on HOT3D and ARCTIC
SucceededMind · x · 2026-11-05
Perceptron launched an egocentric video annotation system that breaks clips into step-by-step action timelines, with separate action labels for each hand or gripper — solving the problem of one label per clip losing fine-grained detail.
Key numbers:
- 12–25% better hand-shape reconstruction accuracy than ACE-Ego-Hand on HOT3D and ARCTIC
- 0.583 segment F1 on WGO-Bench, claimed as the highest reported score
- EgoHome, a new benchmark focused on subtasks shorter than one second
The app is built with Svelte and uses self-service APIs for uploading videos, tracking annotation jobs, and downloading labels, supporting up to 1,000 videos per batch — aimed at robotics teams needing detailed hand annotations at volume.
More from Embodied
- CoRL paper GUILD: actively guiding robot data collection toward useful auxiliary tasks — RobobertoMM · 2026-11-05
- TAMPAgent at CoRL: coding agents + TAMP planning for zero-shot long-horizon mobile manipulation — dhadfieldmenell · 2026-11-05
- Raven Prism launches: a full Linux computer in glasses, controlled by eyes and voice — Scobleizer · 2026-11-05
- Johns Hopkins hosts 'Embodied AI in Healthcare' symposium on December 14 — mdredze · 2026-11-05
- How Hugging Face rebuilt robotics open source in 3 years, from LeRobot to RoboColosseum — DJiafei · 2026-11-05
- The CERN analogy fits robot evaluation: shared hardware, protocols and published failures — DJiafei · 2026-11-05