For people with diabetes, estimating the amount of carbohydrates in a meal is difficult. However, this information is crucial for the correct insulin dosage.

The Diabetes Center Berne is researching a solution to this problem with its startup CarbVis. A smartphone app automatically recognizes the components of a meal and reliably estimates the amount of carbohydrate.

We supported CarbVis in developing suitable algorithms and a computer vision pipeline for automatic food recognition and volume estimation. For better verifiability of the results, we generated a vast database of synthetically generated images of realistic food components.

Skills and Technologies

  • Software development in Python
  • Computer vision
  • Machine learning
Photo by CA Creative on Unsplash