WIth all the polygonization options (#210 should add them) it'd be nice to be able to try them out without having to rerun inference. Once we have #103 then it'd be good to let users run polygonization without having to re-run inference, to see different results from different choices.
