GanHand: Predicting Human Grasp Affordances in Multi-Object Scenes | Publicación