Skip to content

Dataset · openfoodfacts

Annotated nutrition facts tables from food packaging

openfoodfacts/nutrient-detection-layout

About 3 thousand label photos with token annotations for extracting product composition data.

What's inside

The dataset contains images of nutrition tables with markup for the token-classification task, combining images and text. Inside are about 3,000 examples: roughly 2,800 in the training set and 199 in the test set. Images are taken from the open Open Food Facts database for training models as part of the Nutrisight project.

What you can do with it

  • Training LayoutLM-like models for parsing label tables
  • Creating OCR pipelines for automatic recognition of nutritional value and composition
  • Benchmarking structured data extraction systems from documents
Tasks
Token classification
Size
1K–10K rows
Downloads this month
281
Likes
5

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("openfoodfacts/nutrient-detection-layout")