Having the ability to get bounding boxes of images would make it useful for a much wider variety of use cases, where reconstruction of the original or access to all the information is needed.

→ View original post on X — @jeremyphoward