The backend doesn't know that different frames in an icon can have different visual extents because of transparency.

But having the backend aware of the empty space in an icon could deal with that.

Hopefully, it might be possible in the future to account for this sort of thing by having the frontend scan icon bounds.