The COTe score: A decomposable framework for evaluating Document Layout Analysis models

Traditional metrics fail for Document Layout Analysis because they treat images as simple 2D projections. This paper introduces the Structural Semantic Unit (SSU) and the COTe score, a decomposable metric that measures parsing quality by tracking Coverage, Overlap, Trespass, and Excess. This framework offers a more robust evaluation of DLA models by focusing on structural and semantic accuracy rather than just pixel overlap.

Open the full episode

More from AI Radio