Open Food Facts

The open, collaborative database of food products.

AI inside the tool Food and farming Health Digital commons

Contributors around the world photograph packaging and enter ingredients, nutrition facts, labels and origins. The data is published under an open licence and reused by apps, researchers and public services.

And where does AI come in?

Machine learning tools read packaging photos and suggest information (ingredients, labels, categories), which volunteers then check. The AI prepares the work; a person makes the call.

The AI label

Model or technique: YOLOv8 object detectors (logos, Nutri-Score, nutrition tables), CLIP for logos, rules on extracted text.
Model published: Yes
Where it runs: Project servers at OVHcloud; text in photos is read by Google Cloud Vision.
Data: Contributors' photos and product data (database under ODbL, images under CC BY-SA); annotated sets published on Hugging Face.
Role of people: Contributors validate suggestions; only the most certain ones are applied automatically.
Frugality: Not filled in

“Not filled in”: MAUB could not check it with the project. Know more? Write to us.

Something wrong or out of date? Write to contact@subalpine.it.