Bring minimum on-device moderation using MLKit

[teolemon]

What

• We currently perform OCR, NSFW detection server-side.
• We could do some of that work on-device for high-end devices

Why

• prevent bad images from reaching the server
• let malicious actors know quickly and visibly that we have strong systems in places.

How

• https://developers.google.com/ml-kit/vision/image-labeling?hl=fr
• https://developers.google.com/ml-kit/vision/face-detection?hl=fr
• It seems that we can add custom Tensorflow models, which means we could label Nutrition facts, Packaging instruction, Front photo on device, in addition to NSFW and selfie detection

Part of

• #4027

[RitikaxShakya]

@teolemon Hello! I am new here. I would like to know if this issue counts in "Good first Issue"?. Because I landed here from this https://github.com/openfoodfacts/smooth-app/issues/525 Issue which have Good first issue tag. Thank you.

[M123-dev]

Heyy @RitikaxShakya that's a hard one.

Don't hesitate to write in an issue that you want to work on, regardless if a good first issue or not. But it should preferably be something where the instructions are clearer and the issue is quicker to fix, to get to know the project.

Here, if we decide that we want it we need benchmarks and changes in a lot of places.

Looking through #525 is indeed a good idea

[RitikaxShakya]

@M123-dev Thank you for replying, I agree with you picking issues with instructions and quicker to fix at first is good to go for starting. But I looked into https://github.com/openfoodfacts/smooth-app/issues/525 issue and unable to find machine learning issues that have clear instruction or something that I can work on for starting.