Pixels to people, and back

Javier Romero (Amazon Barcelona, Spain)


We live in a world with an increasing digital presence, where most places and buildings are mapped. This trend is only growing, but there is a crucial element that remains digitally inaccessible: people. In this talk I show how our team uses high quality data to create generative human body models. The purpose of these current models is twofold: to infer bodies from images and video, and to generate synthetic data that can be used to train discriminative methods later used for inference. But people are more than bodies; in the last part of the talk I'll discuss additions to the body model like clothing, faces and hands.