FoodieCal: a convolutional neural network based food detection and calorie estimation system

Abstract

According to recent studies across the world, we can see that a healthy diet is the key to having a sound health and body. People nowadays are more concerned with their diets than ever before. With the advancement of science, it is now viable to construct a unique food identification system for keeping track of day to day calorie intake. However, building this kind of system creates several complications on constructing and implementing the model. In our paper, we have developed a new neural network based model which will predict the food items from a given image and show us the estimated calorie of the detected food items. In order to achieve our goal, we have prepared a dataset of around 23000 images for 23 different food categories. Initially, we have developed a single food detection system combining CNN max pooling and ResNet. From our experimentation, we have achieved 93.33% accuracy in this case. Furthermore, we have also taken a step forward to build a system which can detect multiple foods by training CNN with features extracted by Inception V3. We have achieved 89.48% accuracy for this model and we deployed both of our systems on a webpage. The user has to upload an image of food item in the webpage and our system will predict the food item along with the estimated calories in real time.