General Egocentric Video
- 20,000+
- Unique tasks
- ~95%
- Hand visibility
- 1080p+ / 4K
- Resolution
- 30 fps
- Frame rate
We typically reply within 24 hours
First-person, head-mounted-style recordings spanning 20,000+ unique tasks across households, factories, shops, and many more environments, with synchronised IMU data, 95% hand visibility, and a clear field-of-view.
First-person video of real activity captured at the wearer's point of view across 20,000+ distinct tasks: household chores, light industrial and factory operations, retail and shopfloor workflows, hospitality, repair, mobility, and outdoor errands. Every recording is paired with synchronised IMU data, maintains roughly 95% hand visibility, and is shot with a clear, unobstructed field-of-view so hand-object dynamics and scene context survive end to end.
Every clip is collected from paid contributors with explicit consent, scene-level provenance attached, and any non-consenting bystanders blurred during QA before delivery.
Highlights
- 20,000+ unique tasks captured across households, factories, shops, and many more real-world environments
- Every clip synchronised with IMU data and shot with a clear, unobstructed field-of-view
- Roughly 95% hand visibility on average, so hand-object dynamics survive the full duration of the activity
- Authentic first-person footage from paid contributors, not staged actors or scripted demos
- Scene-level consent, bystander face blurring, and commercial-use rights confirmed up-front
Environment coverage
20,000+ unique tasks captured across households, factories, shops, and many more environments. Coverage extends to bespoke scenes, regions, and activity targets on request.
Capture and format
Continuous high-resolution first-person video at 1920×1080 or higher and 30+ fps from head- or chest-mounted rigs, with synchronised IMU (accelerometer and gyroscope) on every clip. Recordings maintain roughly 95% hand visibility, hold a clear unobstructed field-of-view, and span the full lifespan of a task rather than cherry-picked moments.
Annotations
Layered annotation: per-clip scene, location, lighting, and motion metadata as standard, with optional activity and sub-step labels, on-frame object inventory, hand-object interaction events, and short-form first-person narration available on request.
Provenance
- Paid contributors with signed scene-level consent
- Bystander faces blurred during QA before delivery
- No third-party copyrighted content (TVs and monitors are masked)
- Per-clip audit trail and licensable commercial usage rights
Use cases
- Egocentric perception, action recognition, and embodied robotics policies
- Vision-language grounding with first-person narration and IMU context
- AR and VR contextual assistants and procedural-task models
- Hand-object interaction and affordance learning
