I’m interested in computer vision, machine learning, optimization, graphics and robotics.
The COTe score: A decomposable framework for evaluating Document Layout Analysis models
arXiv Preprint, 2026
We introduce the Structural Semantic Unit (SSU), a relational labelling approach that shifts the focus from physical to semantic structure, and the Coverage, Overlap, Trespass, and Excess (COTe) score, a decomposable metric for measuring page parsing quality.
The Character Error Vector: Decomposable errors for page-level OCR evaluation
arXiv Preprint, 2026
Under page-parsing errors, traditional Character Error Rate (CER) becomes undefined. We introduce the Character Error Vector (CEV), a bag-of-characters evaluator that can be decomposed into parsing, OCR, and interaction error components. This allows practitioners to focus on the part of the pipeline with the greatest impact on extraction quality.