Vitadio is a lifestyle management app for people with type 2 diabetes. Through educational modules, therapeutic background and monitoring tools, it helps its users establish healthy routines and daily habits. In the app, users can record their meals and create an overview of their diet. We've added to this functionality too - through food recognition in the uploaded photo.
How does it all work? The user takes a photo of the food and our model automatically identifies the most likely ingredients that make up the dish. There is no need to write down the food manually. It is all about simplifying and streamlining the whole process of entering and describing food. Vitadio then delivers feedback to the user based on just the recognized and entered ingredients.
In the model, we combine two processes: first, the individual objects in the photo are detected and then each object is classified, i.e., mapped to a predefined food label. For this, we initially use the already pre-trained Yolo model on the database, which we have further trained according to specific requirements.
"Enough data is needed to train the model correctly. We are currently working on how to automate the model training in the future with other foods added by users, such as tofu," adds data specialist Andrea Rozhoňová, who is working on the project.
Vitadio has provided us with thousands of images to train the model, and currently, we can reliably identify almost 50 types of food, covering about 50% of all the most commonly labeled foods by users.
"Automatic food recognition in photos is our long-term project to make it more convenient and easier to record patients' diets. We combine the recognition model on our side with a proprietary algorithm that provides personalized feedback to patients," adds Jan Mareš for Vitadio.
Perhaps you also work with images or text material and would like to automate processes when processing them. Get in touch.
Vitadio is a mobile app to support self-management of type 2 diabetes. Vitadio is a CE marked medical device.