Do you have the scene descriptor in the pre-generated dataset provided? Such that we can reconstruct the scene while leverage the annotated grasping pose? Thanks!