I think the idea was that there could be more specialized approaches that are optimized towards field boundaries or even how this dataset is done. Indeed there's likely even approaches of trying to get polygons directly from inference / using ML in the generation of the polygons. I don't think we need to do anything now, just put an issue to say that we may explore others in the future. @calebrob6 may have some specific ideas here.
@cholmes wrote: