Anti-spoof-mn3

[Soheb]

Hi,

I’m evaluating anti-spoof-mn3 for a real-time presentation attack detection use case in an Android app and will be deploy on client devices. We will not modify the model only run inference on-device.

Could you please confirm the following?

1. License & Commercial Use
• Is the anti-spoof-mn3 ONNX model (weights) licensed in a way that permits commercial use in a closed-source Android app (e.g., MIT as stated in the repo/docs)?
• Are there any additional restrictions (redistribution terms, attribution wording, NOTICE requirements) when bundling the unmodified weights in an app?
2. Weights vs. Datasets
• Were the released weights trained on datasets such as CelebA-Spoof (or others) that impose downstream usage restrictions?
• If so, do those dataset licenses affect commercial redistribution/use of the resulting weights, or are the weights explicitly cleared for commercial use?
3. Attribution & Notices
• What exact attribution text and license files should we include in our in-app OSS notices?
• Do you require a specific citation (paper/repo)?
4. Compliance Extras (if any)
• Any guidance on naming/branding usage, or patent considerations we should be aware of?

Thanks,

Soheb.

[Wan_Intel]

Hi Soheb,

Thank you for reaching out to us.

Let me check with relevant team, and we will provide an update here at the earliest.

Regards,

Wan

[Zulkifli_Intel]

Hi Soheb,

Thank you for your patience.

1. License & Commercial Use

•  Is the anti-spoof-mn3 ONNX model (weights) licensed in a way that permits commercial use in a closed-source Android app (MIT License in Legal Information)?

The model is released under the MIT License (stated in Legal Information).MIT License permits commercial use, redistribution, modification, and sublicensing.

• Are there any additional restrictions (redistribution terms, attribution wording, NOTICE requirements) when bundling the unmodified weights in an app?

Usual obligations: keep the copyright & license notice in redistributions. No restriction against closed-source Android apps.

2. Weights vs. Datasets

• Were the released weights trained on datasets such as CelebA-Spoof (or others) that impose downstream usage restrictions?

anti-spoof-mn3 weights were trained on CelebA-Spoof and other open datasets (documented in model.yml).

• If so, do those datasets licenses affect commercial redistribution/use of the resulting weights, or are the weights explicitly cleared for commercial use?

ANS: CelebA-Spoof itself has a non-commercial clause, but OpenVINO clarifies that released weights are cleared for commercial use (Intel handles licensing due diligence).

3. Attribution & Notices

• What exact attribution text and license files should we include in our in-app OSS notices?

You need to include:

MIT License text (from the repo LICENSE file).

Attribution: “anti-spoof-mn3 model from Open Model Zoo (OpenVINO Toolkit)”

• Do you require a specific citation (paper/repo)?

Citation (recommended but not strictly required): refer to the original paper + OpenVINO repo link.

4. Compliance Extras (if any)

• Any guidance on naming/branding usage, or patent considerations we should be aware of?

No naming/branding restrictions, but don’t use Intel/OpenVINO trademarks as product branding.

No patent clauses are listed in MIT; you can use it at your own risk.

I hope these answers are enough for you.

Regards,

Zul