Human demonstrations · Meta Reality Labs / Yale University
SHOW3D
Synchronized egocentric and external views of human hand–object interactions with 3D annotations.
Release / update
Hand pose v3 release update, documented in the publisher card.
What is included?
- Data type
- Human demonstrations
- Tasks
- Hand-object interaction · Multi-view perception
- Modalities
- Monochrome video · Hand pose · Object pose · Language · Camera calibration
- Embodiment
- Human participants with a headset and a mobile external-camera rig
Reported scale
4,276,772 synchronized multi-view framesPublisher release-card frame count. Frames are not individual camera images or robot trajectories. The card separately reports 20 recording hours.
Access & license
The official Hugging Face card links scene files, annotation versions, split indexes and the dataset API.
Before you use this dataset
Human demonstrations, not robot actions. Camera and annotation availability vary by scene; inspect presence and validity flags. The calibration world frame moves with the capture rig. Commercial use is not permitted under the stated license.
Publisher sources checked 2026-10-05. Listing review checks the documentation, not downloaded data quality. Always check the exact release and terms.