Image-Based Body Shape Estimation to Detect Malnutrition

The detection of malnutrition in children contributes to the United Nations' second Sustainable Development Goal (SDG2): Zero Hunger. One of SDG2's indicators is the prevalence of malnutrition among children under the age of five. Certain body measures such as stature (height) and head circumference are typically used to assess growth and malnutrition in children. In this paper we examine the feasibility of using convolutional neural networks (CNNs) to infer body shape directly from images. We aim to (i) predict three body measurements: height, head circumference and waist circumference, and, (ii) using a parameterised body model, predict the body-shape parameters from images. We created a multi-view collection of images of human bodies based on the CAESAR and AGORA datasets. Our predictions of the three body measurements are competitive with those obtained in a recent study for stature and head circumference, but not for waist circumference. Our predictions of the body-shape parameters, yields reasonable estimates of the body shape parameters, that seem to be hampered by pose and size variations. Our findings lead us to conclude that imagebased assessment of body shape seems feasible. Further work is needed to assess the potential of parameterised body models and the generalisation to in-the-wild assessment of child malnourishment.