Dish-ID: A neural-based method for ingredient extraction and further recipe suggestion

摘要

The paper presents a method for meal recognition, ingredient extraction and recipe suggestion in the Russian language. The proposed algorithm consists of several consecutive stages. On the first stage the model extracts a list of ingredients from a photo of the dish, based on which recipes on the second stage are selected. Two ingredient extraction architectures were tested for the first stage and three recipe matching methods for recipe suggestion are proposed. In addition, the algorithm was incorporated into the telegram-bot which provides friendly user experience. Source code is at https://github.com/Alenushldish_id_sirius.