What can I work on ? (off-category-classification)

[teolemon]

Who for

• Contributors, Users

What

• Setup

  • install training project #3
  • run the models and see if everything works #4
  • Make a dockerfile / docker-compose that train the model #5

• Data sets

  • decide on formats for generated datasets #8
  • Get familiar with the API #9
  • Create new dataset #10
  • make a standard training / test / verification data set #13

• Explore the existing data in notebooks

  • Explore feature sets in notebooks #14
  • Explore Robotoff's existing data #15

• Adding features to model

  • Use Pictures as a complementary feature for off-category-classification #6
  • Use NLP on OCR as a complementary feature for off-category-classification #7
  • Use Nutrition as a complementary feature for off-category-classification #11
  • Integrate all features in a unique model #12 (THE BIG TARGET)
  • Investigate the use of logos and labels as a feature for the category classifier #18

• Improvements in applications

  • install Robotoff (see https://github.com/openfoodfacts/robotoff/blob/master/doc/introduction/contributing.md#get-started)
    - Improve documentation

• Monitoring

  • see Robotoff quality monitoring robotoff#660 and sub issues

• Do not predict categories that are not taxonomized anymore

Why

• Compute more Eco-Score and Nutri-Score, more reliably