Face To BMI: A Deep Learning Based Approach for Computing BMI from Face

Abstract

Body Mass Index (BMI) is a measure of how healthy a person is with respect to their body weight. BMI has shown a correlation with various factors like physical health, mental health, popularity. BMI calculation often requires accurate height and weight, which would take manual work to measure. Largescale automation of BMI calculation can be utilized for analyzing various aspects of society and can be used by governments and companies to make effective decisions. Previous works have used only geometric facial features discarding other information, or a data-driven deep learning-based approach in which the amount of data becomes a bottleneck. We used the state of the art pre-trained models such as Inception-v3, VGG-Faces, VGG19, Xception and fine-tuned them on the comparatively large public dataset with discriminative learning. We used the larger Illinois DOC labeled faces dataset for training and Arrest Records, VIP_attribute for evaluation purposes.