← 研踪 PaperTrailarXiv:2607.00886 · v2
From Formulas to Protocols: A Compositional View of Binary Target Segmentation Metrics
首次提交 2026-07-01 20:53
版本更新 2026-11-17 12:17
雷达观测 2026-11-18 19:43
摘要
Evaluation metrics are central to binary target segmentation because they determine how progress is measured, compared, and interpreted. In this paper, target denotes the task-defined positive region to be segmented rather than a generic foreground object. It may be salient, camouflaged, transparent, glass-like, mirror-like, shadow-like, lesion-like, or defined by other application-specific semantics. We treat metrics as compositions of modular design choices rather than isolated formulas. The proposed framework decomposes each metric into the five stages of prediction representation, target extraction, target matching, score computation, and metric reporting. Analyzing representative metrics by stage keeps their assumptions visible and shows how newer metrics address specific limits in earlier protocols. Controlled examples show how protocol choices affect evaluation outcomes and how existing mechanisms can be recombined for specific evaluation needs. We then discuss the resulting design space and its implications for task-aware evaluation protocols. Reference code is available at https://github.com/lartpang/PySODMetrics
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-17 12:17 · 最新观测版本
From Formulas to Protocols: A Compositional View of Binary Target Segmentation Metrics
Evaluation metrics are central to binary target segmentation because they determine how progress is measured, compared, and interpreted. In this paper, target denotes the task-defined positive region to be segmented rather than a generic foreground object. It may be salient, camouflaged, transparent, glass-like, mirror-like, shadow-like, lesion-like, or defined by other application-specific semantics. We treat metrics as compositions of modular design choices rather than isolated formulas. The proposed framework decomposes each metric into the five stages of prediction representation, target extraction, target matching, score computation, and metric reporting. Analyzing representative metrics by stage keeps their assumptions visible and shows how newer metrics address specific limits in earlier protocols. Controlled examples show how protocol choices affect evaluation outcomes and how existing mechanisms can be recombined for specific evaluation needs. We then discuss the resulting design space and its implications for task-aware evaluation protocols. Reference code is available at https://github.com/lartpang/PySODMetrics
查看此版本解读 →查看此版本原文 ↗