Licenses¶
Py-Feat's original software is MIT-licensed, including commercial use. Pretrained models and research datasets have separate terms. Choose the notice for the product and model family you use:
| Product | Original software | Models and data |
|---|---|---|
| Py-Feat v1.0 / Detectorv1 | MIT | Separate notices for each component and fitted artifact. |
| Py-Feat v2.0 / Detectorv2 | MIT | Multitask checkpoint, pretrained backbone, teachers, and training data have separate terms. |
| PyFeat-Live | MIT | The selected detector and optional model assets keep their own terms. |
“V1.0” and “v2.0” here identify the classic and multitask model families, not Python package version numbers. See the dataset register for training, model-selection, and evaluation roles and the evidence supporting each entry.
For a future model intended for commercial users, see the commercial retraining table. It distinguishes commercial training, public weight sharing, and unresolved permissions.
The software license does not establish unrestricted commercial use or redistribution of every pretrained asset. Some assets have explicit noncommercial terms; others have unresolved upstream grants or dataset-agreement scope. An unresolved permission is not filled by adding an MIT or research-only label. Existing valid grants are not revoked by these notices.
The repository MIT License and third-party notices should be preserved in distributions. Signed data agreements and private correspondence are retained outside the public repository. This page was reviewed October 8, 2026.
Commercial-use inquiries¶
The current multitask weights are not offered as commercially cleared weights. Py-Feat does not broker third-party permissions or issue a commercial sublicense that overrides another provider's restrictions. This does not revoke valid existing grants or restrict MIT-licensed code.
Commercial users can use independently cleared weights, or assess the exact checkpoint and contact the relevant rights holders directly for any necessary permissions. The v2 notice explains the scope; the dataset register links provider sources. A routine dataset access approval is not necessarily commercial checkpoint permission. User-obtained permissions also do not automatically resolve the distributor's separate obligations.