EgoDex is an excellent public dataset. This page compares it factually with 10KH-Ego Stereo so buyers can decide which fits their use; the two often complement each other (public data for pre-training breadth, licensed data for commercial training and target scenes).
| EgoDex | 10KH-Ego Stereo | |
|---|---|---|
| Publisher | Apple | 10K HoursData |
| Scale | 829 hours, 338,000 episodes, 194 tabletop tasks, 1080p 30 Hz | an existing library of 400k+ monocular hours across about 40 real scenes, with a curated subset of thousands of hours |
| Capture | Apple Vision Pro with ARKit 3D hand/finger and upper-body pose | Head-mounted capture; see the hardware page |
| License | CC BY-NC-ND (non-commercial, no derivatives) | Commercial license per project |
| Custom collection | No | Yes — 100k+ effective hours per month of new collection capacity (scenes with strict requirements noticeably reduce what can be delivered in a given month) |
Sources: github.com/apple/ml-egodex, arXiv:2505.11709.
See also: how to choose a supplier · 10KH-Ego Stereo
FAQ
Can EgoDex be used for commercial robot training?
No — EgoDex is CC BY-NC-ND, which excludes commercial use.
When should I choose 10KH-Ego Stereo over EgoDex?
When you need commercially licensed data, specific target scenes, or new collection to your spec.
Where can I see a sample?
On Hugging Face.