Ego-Exo4D-HM: SMPL-H reconstructions for 523 hours of egocentric-exocentric video, open-sourced
geopavlakos · x · 2026-10-02
Researchers have released Ego-Exo4D-HM, a large-scale 4D human motion reconstruction dataset built on Ego-Exo4D:
- 523 hours of egocentric and exocentric video across 2500+ captures and 8 activity types
- Provides SMPL-H reconstructions (body + hands) for the camera wearer, produced via multi-view optimization
- Data on Hugging Face plus open-source pipeline: environment setup, downloading, batch processing/resuming, npz format docs, and mesh rendering
- Includes an interactive viewer for browsing the data
A ready-to-use resource for embodied AI, motion capture, and human-object interaction research.
More from Embodied
- Agility's Digit runs end-to-end autonomous whole-body manipulation for ~12 hours at IROS — chris_j_paxton · 2026-10-02
- Tesla FSD slams the brakes just in time — driver in the next lane never noticed — xiaosun86 · 2026-10-02
- Atech's modular hardware kit is vibe-codeable with Claude, no wiring required — Paimaamu · 2026-10-02
- Figure's data collection project Index crosses 1 million signed-up users — adcock_brett · 2026-10-02
- DATAFARM: aligning TAMP demos to VLA pretraining distribution unlocks planning plus dexterity — tomssilver · 2026-10-02
- Wearable AI lets you capture thoughts by voice anywhere you go — broodsugar · 2026-10-02