← Use cases
Use case · Action recognition
Egocentric data for action recognition.
Densely-labeled first-person clips of real activities — the ground truth for models that understand what a person is doing.
The need
Recognizing fine-grained human actions from the first-person view — cooking steps, repairs, assembly, retail tasks — requires egocentric clips with clear temporal boundaries and consistent verb–noun labels, including long untrimmed video for real-world robustness.
What we provide
- Fine-grained activity and procedure clips
- Multi-step, sequential tasks
- Long, untrimmed first-person video
- Multiple domains — kitchen, repair, retail, and more
How it's collected
- Scripted procedures plus natural activity
- Consented contributors
- Consistent task design for clean labels
How it's delivered
- Temporal action segmentation and verb–noun labels
- Reviewed, model-ready annotations
- Documented provenance
More use cases
Need Action recognition data?
Tell us the tasks and environments — we'll scope the capture.