VisCurate: Evaluating Behavioral Equivalence of Visual Skills

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.30405 (cs) [Submitted on 21 Oct 2026] Title: VisCurate: Evaluating Behavioral Equivalence of Visual Skills Authors: Md Farhan Ishraq, Md Farhan Ishmam, Reaz Hassan Joarder, Md Hasanul Kabir.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.30405 (cs) [提交于 2026 年 10 月 21 日] 标题：VisCurate：评估视觉技能的行为等价性 作者：Md Farhan Ishraq, Md Farhan Ishmam, Reaz Hassan Joarder, Md Hasanul Kabir。

Abstract: Modern vision-based agents have shifted toward skill-based problem-solving, where a skill is an executable image-to-image operation, e.g., cropping, thresholding, or edge detection. As these skill libraries grow, they accumulate duplicated, mislabeled, and defective skills. The signals normally used to curate them (a skill’s name, description, or source code) say little about how the skill visually changes the image.

摘要：现代基于视觉的智能体已转向基于技能的问题解决模式，其中技能是一种可执行的图像到图像的操作，例如裁剪、阈值处理或边缘检测。随着这些技能库的增长，它们积累了重复、标记错误和有缺陷的技能。通常用于整理这些技能的信号（技能名称、描述或源代码）对于技能如何从视觉上改变图像几乎没有提供信息。

We present VisCurate, a framework and benchmark that curates visual skill libraries from observed execution rather than from text. VisCurate runs each candidate skill pair on a shared probe battery and assigns one of six behavioral relations, each tied to a library edit that is safe to make.

我们提出了 VisCurate，这是一个通过观察执行而非文本来整理视觉技能库的框架和基准。VisCurate 在共享的探测电池上运行每一对候选技能，并分配六种行为关系之一，每种关系都与一个可以安全进行的库编辑相关联。

The benchmark injects seven defect types into a clean library at controlled rates, producing 300 reproducible instances with ground-truth relations and repairs. While text baselines accept up to five of six false merges on hard negatives, our verifier accepts none and recovers 51.2% of the pairs that are truly mergeable.

该基准以受控速率向干净的库中注入七种缺陷类型，产生了 300 个具有地面实况关系和修复方案的可复现实例。虽然文本基准在处理困难负样本时会接受六个错误合并中的多达五个，但我们的验证器一个也不接受，并恢复了 51.2% 真正可合并的对。

On two real, uncurated libraries, VisCurate finds that 92% of the skills a package shares with its downstream copy are redundant, while execution confirms only 10 of 17 published equivalence claims. The curation problem also remains open, as the strongest of the nine automated curators recovers only about half of the repairs. We emphasize that visual skill libraries should be curated by behavior, not by description.

在两个真实的、未经整理的库上，VisCurate 发现一个包与其下游副本共享的技能中有 92% 是冗余的，而执行结果仅证实了 17 个已发布等价声明中的 10 个。整理问题仍然悬而未决，因为九个自动化整理器中最强的一个仅恢复了大约一半的修复方案。我们强调，视觉技能库应根据行为而非描述进行整理。