Difference between revisions of "DataForGood-2022"

Revision as of 14:39, 15 March 2022

DataForGood is an association that mobilize tech to help cytizen projects.

Main focus this year is on using machine learning to automatically categorize products so as to enable computing Nutri-Score and Éco-Score.

Status: started

Expected outcomes: Have a new machine learning model, ready to deploy in production:

That use more features to predict category: OCR data, nutrional information, eventually images or any feature
Precision should be very high to be able to apply category automatically but the model should use it's confidence to ask for user when needed

Impact: augment massively the number of Eco-Score and Nutri-Score we are able to provide on the platform.

Categorization is also important in many way, for example to compare nutrition data with other products of same category (rank among category).

Timeline:

The weekly meeting for this project is every wednesday 20.00. We have a meeting room at https://meet.jit.si/DFG-OPENFOODFACTS

You should also join data-for-good slack. To be invited, use https://dataforgood.fr/join/

Main repos

robotoff: Our program that handles predictions / insights: good to read : https://openfoodfacts.github.io/robotoff/introduction/architecture/
robotoff-ann is a complement to robotoff that focus on logo detection and embedding (we use nearest neighbors to classify logos)
last training of category model (using only title and ingredients): https://github.com/kulizhsy/off-category-classification
All trained models are published as "releases" on https://github.com/openfoodfacts/openfoodfacts-ai
openfoodfacts API documentation