Guide
How to license training data for commercial models
Short answer
A commercial model needs training data whose license permits commercial use and whose consent is documented per participant. Many public egocentric datasets are research-only. Before training, confirm the license grants commercial rights, that consent is evidenced by signed releases, and that a data card records jurisdictions of capture.
What to confirm before you train
- The license explicitly permits commercial use, not research-only
- Rights are perpetual and buyer-owned, not a revocable subscription
- Every participant signed a release granting commercial training rights before capture
- A data card records the consent basis and jurisdictions of capture
- Bystanders are de-identified irreversibly at the pixel level
Why does research-only data become a liability?
Training on data with murky consent or a research-only license is a legal and reputational risk that surfaces after deployment, when the model is already in a product. A verifiable consent chain and a commercial license are what let a commercial team actually ship a model trained on human demonstration data.
Firsthand ships a perpetual, buyer-owned commercial license with consent artefacts and a data card per batch. Compare public dataset terms on /compare/public-dataset-licenses.
Check it against the sample pack.
40 episodes across 4 environments, delivered in the exact schema these guides describe.