Face To BMI: A Deep Learning Based Approach For Computing BMI From Face

Abstract

Body mass index (BMI) is a measure of a person's health in relation to their body weight. BMI has been shown to correlate with various factors such as physical health, mental health, and prevalence. Calculating BMI often requires exact height and weight, which will require manual work to measure. Large scale automation of BMI calculation can be used to analyze different aspects of society and can be used by governments and businesses for to make effective decisions. Previous work used only geometric facial features that removed other information or the data-driven deep learning approach where the amount of data became a bottleneck. We used pre-trained modern models such as Inception-v3, VGG-Faces, VGG19, Xception and refined them on a relatively large public dataset with discriminant learning. We used the largest dataset of faces labeled Illinois DOC for training and Capturing Profile, VIP attribute for evaluation purposes.