Thank you so much for providing the nice dataset.

During some exploratory analysis, I've noticed that the partitioning of the 24 groups are not disjoint in the sense that a couple of classes might have been mistakenly grouped into two different parent classes by accident: classes bandage and nail belong to both Tools and Office, in Figure 2 and the Appendix.

I understand that this is a rather minor issue and does not have an impact on model training and evaluation. I'm wondering how would you group bandage and nail if they were to be partitioned into disjoint divisions/subsets?

Thank you again!