AI Estimates Weight and Height from Single Wild Images.

A person's physical attributes like weight and height are crucial health indicators, often summarized by Body Mass Index (BMI). Automatically estimating these from a single image captured "in the wild" presents significant challenges due to diverse human poses, camera angles, appearances, and distracting backgrounds. This study investigates the performance of deep neural networks using various modalities, including RGB, depth-maps, pose-affinity maps, and edge-maps, to predict BMI, weight, and height. To facilitate this research, a new dataset of 6105 images, complete with ground truth labels for height, weight, and BMI, has been created. This dataset, collected from social media, features diverse ethnicities, age groups, genders, poses (frontal, back, side, mirror selfies), and backgrounds, including images with partial occlusions. Extensive experiments using different CNN backbones (VGG, Densenet, ResNet) show that full-body images consistently produce superior results compared to half-body or face-only images. This work contributes both a novel dataset and insights into robust single-image anthropometric estimation.