Deep learning techniques for visual food recognition on a mobile app

Abstract

The paper provides an efficient solution to implement a mobile application for food recognition using Convolutional Neural Networks (CNNs). Different CNNs architectures have been trained and tested on two datasets available in literature and the best one in terms of accuracy has been chosen. Since our CNN runs on a mobile phone, efficiency measurements have also taken into account both in terms of memory and computational requirements. The mobile application has been implemented relying on RenderScript and the weights of every layer have been serialized in different files stored in the mobile phone memory. Extensive experiments have been carried out to choose the optimal configuration and tuning parameters.