Food Recognition with CNN

My colleague Tomas Costa and I developed a Machine Lerning module to recognize and label different types of food.

Idea

The idea came from our need for such a module in the context of a larger project in informatics (still in development), along with the idea of developing a project involving neural networks for our Machine Learning course.

Development

After trying different methods, we opted by using Convolutional Neural Networks (CNNs).

You can check out the code and other information at this repo.

We have developed everything using Google Colab, which allowed us to run some relatively compute intensive tasks on a remote server for longer times, without needing our laptops.

Results

We weren’t able to get a perfect result. However, given the complexity of the problem and our available resources, we are happy with the relatively good results we accomplished, even with a smaller dataset.