Revision as of 13:39, 28 January 2022

Machine learning models and "AI" are used in Open Food Facts to extract information from product images and to perform predictions, such as the product category.

Principles

AI is there to help accelerate our transparency mission by augmenting Open Food Facts contributors, letting them focus on value-added tasks and increasing their personal impact.
We can use AI and non-AI techniques together and blend them, so that when one of the technique fails for some reason, the other one can save the day.
Even if a model does not have perfect results, at Open Food Facts scale, it might save contribution time and increase food transparency. Furthermore, it's possible to leverage the community to validate guesses and improve the models over time.

Projects

Robotoff

Robotoff is Open Food Facts "AI" service: it takes care of calling the ML models and storing all predictions. Robotoff documentation describes in details how Robotoff works.

Depending on its confidence, each prediction is either applied automatically or require a manual validation. Hunger Games is our in-house validation tool.

All machine learning models used in production are stored on Github as releases, in the robotoff-models repository.

Hunger Games

Hunger Games is our collaborative and playful annotation engine. It allows you to annotate in bulk logos and labels, to mass complete nutrition tables and ingredient lists, and to answer Robotoff questions in vast numbers.

Models

A list of our AI projects (current and future) can be found on Github: openfoodfacts-ai.

Nutri-Score prediction

Our classic model that's able to predict the letter of the Nutri-Score on the packaging is soon to be obsoleted by our more generic logo detection model.

Category prediction

Logos and labels detection

Nutrition facts table data extraction

The Nutrition facts table data extraction project aims to automatically extract the nutrition facts values from photos of nutrition facts table.

Ingredients spellchecking and correction

How to contribute?

There are many ways to help:

contribute to Robotoff, some issues are tagged as suitable for newcomers
help to solve one of the issue listed in openfoodfacts-ai.

Most of our discussions happen on Slack. Depending on your interests, you can join the following channels:

#robotoff and #hunger-games (robotoff and hunger games channels respectively)
#ai-machinelearning (generic machine learning discussions)
#ai-logo-detection, #ai-imagerecognition, #ai-computervision, #spellcheck

@@ Line 6: / Line 6: @@
 * We can use AI and non-AI techniques together and blend them, so that when one of the technique fails for some reason, the other one can save the day.
 * Even if a model does not have perfect results, at Open Food Facts scale, it might save contribution time and increase food transparency. Furthermore, it's possible to leverage the community to validate guesses and improve the models over time.
 = Projects =
 == Robotoff ==
-[https://github.com/openfoodfacts/robotoff Robotoff] is Open Food Facts "AI" service: it takes care of calling the ML models and storing all predictions. [https://github.com/openfoodfacts/robotoff/wiki Robotoff wiki] describes in details how Robotoff works.
+[https://github.com/openfoodfacts/robotoff Robotoff] is Open Food Facts "AI" service: it takes care of calling the ML models and storing all predictions. [https://openfoodfacts.github.io/robotoff/ Robotoff documentation] describes in details how Robotoff works.
 Depending on its confidence, each prediction is either applied automatically or require a manual validation. [https://hunger.openfoodfacts.org/ Hunger Games] is our in-house validation tool.