Using deep learning for food recognition

Abstract

[en] Image recognition is a very challenging and important problem in the computer vision field. And food image classification is one of the most challenging branches of this field. In real-world scenarios, it is more common for a food image to have more than one food item. As a result, the multi-label classification problem has generated significant interest in recent years. However, multi-label recognition is a much more difficult object recognition task compared to single-label recognition. In this work, we will study the multi-label food recognition problem by using deep learning algorithms, specifically Convolutional Neural Networks. We will show how redefining the loss function as well as augmenting the training dataset can leverage the multi-label food recognition problem. Extensive validation will be presented to show the strengths and limitations of multi-label food recognition.