Questions About MediaPipe Privacy Policy

[Heidi He]

Hi everyone,

I hope you’re all doing well! I'm currently working on an interactive film project called Light Years Apart, which uses the MediaPipe Web API to process biometric data, including audience facial recognition and landmarks. You can check out more about it at www.lya-movie.com.

I have a couple of questions regarding the privacy policy:

1. Does anyone know if the policy about on-device processing still holds true? Specifically, is any data from the Web API sent to Google or external servers?
2. Also, does anyone have information on whether the training data for MediaPipe’s biometric models is inclusive and representative of diverse populations?

If anyone has insights or knows how to get in touch with the MediaPipe team for more detailed info, I’d really appreciate it!

Thanks so much!
Heidi

[Max Chan]

Man I would like to know the same too. Have never considered this side of problem.

Did u find what you need? Thanks

Max

[Heidi He]

I think I replied but maybe the response is not sent correctly so it is not shown in the discussion?

So coping and re-send it again --- 

Hi Max,

Yes, I later emailed them directly and luckily got a response from someone at the Mediapipe's team. I am working on an interactive film that is centered around diversity and inclusivity. So the answers to these questions matters a lot to our choice of ML library in the film. 

This is their response. I find it helpful so also sharing it here:

Hi Heidi!

 

For information on the specific models and how they were trained, the model cards would probably be the best reference available. These are linked to in the "models" section of the task overview. For example, if you wanted more information on our BlazePose GHUM 3D pose landmark detection model, the pose landmarker model information section would be here in our documentation, which links to this model card for that particular model. Our published geographic evaluation results and fairness criteria/metrics/results can be found therein.

 

And yes, on-device processing continues to be a core feature of our web Tasks APIs, so all machine learning is performed directly in the browser, meaning that there is no server involvement in the machine learning process for applications like facial recognition and landmark analysis.

Hope it helps!

Best,

Heidi