nutriscore-object-detection

# Open Food Facts Nutriscore detection dataset This dataset was used to train the Nutri-score object detection model running in production at Open Food Facts. Images were collected from the Open Food Facts database and labeled manually. Just like the [original images](https://world.openfoodfacts.org/data), the images in this dataset are licensed under the Creative Commons Attribution Share Alike license (CC-BY-SA 3.0). ## Fields - `image_id`: Unique identifier for the image, generated from the barcode and the image number. - `image`: Image data. - `width`: Image original width in pixels. - `height`: Image original height in pixels. - `meta`: Additional metadata. - `barcode`: Product barcode. - `off_image_id`: Open Food Facts image number. - `image_url`: URL to the image on the Open Food Facts website. - `objects`: Object detection annotations. - `bbox`: List of bounding boxes in the format (y_min, x_min, y_max, x_max). Coordinates are normalized between 0 and 1, using the top-left corner as the origin. - `category_id`: List of category IDs. - `category_name`: List of category names. ## Versions - `1.0`: Original data used to train the [tf-nutriscore-1.0 model](https://github.com/openfoodfacts/robotoff-models/releases/tag/tf-nutriscore-1.0). - `2.0`: New version of the dataset with improvements over the original version: the bounding boxes are more tightly cropped around the Nutri-score, some labeling errors were corrected, and images for which the model was failing were added to the dataset to improve future versions of the model.