Raphaël Bournhonesque
add v1 of the proof classification model
6e7f851
|
raw
history blame
1.01 kB
metadata
license: agpl-3.0

This model is an image classifier used to predict the proof type of a proof image on Open Prices.

The folling categories are supported:

  • PRICE_TAG: a photo a single price tag in a shop. If several price tags are visible, the image should be classified as SHELF.
  • SHELF: a photo of a shelf with several products and price tags. If only one price tag is visible, the image should be classified as PRICE_TAG.
  • PRODUCT_WITH_PRICE: a single product with a price written (or added) directly on the product.
  • RECEIPT: a photo of a shop receipt.
  • WEB_PRINT: a screenshot of a web page displaying a price/receipt or a photo of a price brochure.
  • OTHER: every other image that doesn't belong to any of the above category.

The model was trained with Ultralytics library. It's was fine-tuned from YOLOv8n model. The training dataset is an annotated subset of the Open Prices dataset.