Commercially licensed egocentric data, built for training and shipping.

Egocentric data is commercially licensed when the rights attached to the footage explicitly permit training and deploying a commercial model on it, not just reviewing or benchmarking against it. That grant has to come from somewhere: either the publisher of a dataset states commercial terms, or the footage was captured specifically for the buyer under a contract that assigns those rights.

On this site, a custom egocentric collection is captured to a written spec and delivered under a perpetual, irrevocable, worldwide license that is non-exclusive by default, with the buyer owning the resulting data. That is a different starting point than footage collected for research publication, where the license usually was never written with commercial deployment in mind.

Public egocentric datasets are frequently released to support academic research and benchmarking, and their license terms are written for that purpose: review, reproduction, and non-commercial evaluation. A research-only license does not typically contemplate a company training a model on the data and shipping it as part of a commercial product.

Checking the specific license text of any public dataset before training on it is the only reliable way to know what it permits. This site links to the official page or license of any public dataset it references, rather than describing someone else’s terms from memory.

A custom egocentric collection ordered through this site ships under a license grant that is perpetual, irrevocable, worldwide, and non-exclusive by default, with Firsthand documenting and warranting the consent chain behind the footage. The data itself is buyer-owned, which is a separate and stronger position than holding a license to data someone else continues to own.

These are general license terms as stated on this site’s provenance and terms pages, not legal advice for a specific contract. Buyers should still have counsel review the actual agreement before training on any dataset.

A commercial license is only trustworthy if the underlying consent is real and documented. Every contributor signs a release before capture, and that documented consent chain is part of what ships with the dataset, not a claim made separately from the footage itself.

Faces, license plates, and screens are detected and irreversibly blurred at ingest, then confirmed by a human reviewer as part of QA before anything is delivered. That anonymization step reduces the privacy exposure of using the footage commercially, on top of the consent and license documentation.

Buyer-owned means the buyer holds the data outright rather than holding a license to data someone else continues to own and could license to others on different terms. Non-exclusive by default describes the baseline grant structure; exclusivity, where wanted, is a separate term to negotiate rather than an assumed default.

Combined with a perpetual, irrevocable, worldwide grant, this removes the two risks that most commonly surface after a model has already been trained and shipped: an expiring license term, and a provider retaining more control over the underlying data than the buyer expected.

Before training on any egocentric dataset — public or custom — check whether the license explicitly permits commercial use, whether it is perpetual or term-limited, whether the grant is exclusive or non-exclusive, and whether consent for the footage is documented or simply asserted. A dataset page that does not answer these questions directly is a reason to ask before committing a training run to it.

For a custom order, those answers live on this site’s provenance and terms pages, and in the documentation delivered alongside the dataset. For a public dataset, check its official license page directly rather than relying on a secondhand description.

Start with a written spec describing the tasks, environments, objects, and streams your model needs. Most orders begin with a Spec Pilot — a small, fast first batch that proves the spec and accept criteria hold up on real footage, with coverage, reject rate, and a real timeline figure, before scaling to the full order.

Where coverage for the requested tasks and regions is already deep, the median time from signed spec to first validated batch is 48 hours. Where it is thin, turnaround runs three to four weeks while recruiting expands in parallel across more regions.