Séminaire IMAGE : « Human Pose Forecasting for intention anticipation », Mathis Dupont

Nous avons le plaisir d’accueillir Mathis Dupont. Il parlera des travaux qu’il a menés lors de son stage de recherche au CEA Paris-Saclay le jeudi 17 octobre à 14h00 en salle de séminaire F-200.

Titre : « Human Pose Forecasting for intention anticipation »

Résumé :
Human Pose Estimation and Human Pose Forecasting are extensively researched fields with diverse applications, including autonomous vehicles, sports analytics, robotics. During this internship, we first focused on Human Pose Estimation and developed a novel transformer architecture designed to aggregate prior bone length information ensuring bone length consistency within a subject. This work also led to the introduction of a new metric for more accurately evaluating bone length consistency. In the second phase of this internship, dedicated to Human Motion Forecasting, we adapted an existing model to explore the effectiveness of Transformers in this domain.