How can i get 3d bbox ground truth of each pose of camera? It seems like benchbot api doesn't support to do so