Affordance-Map: Learning Hidden Human Context in 3d Scenes Through Virtual Human Models