IOS Mobile Application for Food and Location Image Prediction using Convolutional Neural Networks

Machine Learning is a popular research area in software industry alongside with big data, micro services, virtual reality, and augmented reality. With the recent developments in improving computing capacity, deep learning approaches such as Convolutional Neural Networks (CNN) has become the trendiest topic in machine learning for image recognition. In this paper, we have developed an IOS application for food image recognition using modified CNN models. In particular, we developed an IOS mobile application by converting the models to CoreML and then using them within the IOS application for food image recognition. After fine-tuning a pre-trained Google InceptionV3 model, we were able to achieve 82.03% Top-1 accuracy on the test set using a single crop per item. Using 10 crops per example and taking the most frequent predicted class(es), we were able to achieve 86.97% Top-1 Accuracy and 97.42% Top-5 Accuracy.