Since the MAP between the largest and smallest Kitti subset is not significant, another training with even smaller datasets was conducted.