Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincar\'e Ball

arXiv:2404.03778v1 Announce Type: new
Abstract: Hierarchy is a natural representation of semantic taxonomies, including the ones routinely used in image segmentation. Indeed, recent work on semantic segmentation reports improved accuracy from supervised training leveraging hierarchical label structures. Encouraged by these results, we revisit the fundamental assumptions behind that work. We postulate and then empirically verify that the reasons for the observed improvement in segmentation accuracy may be entirely unrelated to the use of the semantic hierarchy. To demonstrate this, we …

abstract accuracy arxiv assumptions bias cs.cv hierarchical image indeed natural reports representation results segmentation semantic supervised training taxonomies training type work