MatchED：基于端到端匹配监督的清晰边缘检测
MatchED: Crisp Edge Detection Using End-to-End, Matching-based Supervision
用可微的一对一匹配替代 NMS 与骨架细化这类不可微后处理，把「训练-测试协议一致性」引入边缘检测；做低层结构预测（边缘、线、骨架、关键点细化）时可直接把该 21K 参数的匹配监督模块挂到现有模型上，实现端到端输出单像素宽结果。
原文依据 MatchED performs one-to-one matching between predicted and ground-truth edges based on spatial distance and confidence
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论生成清晰（即单像素宽）的边缘图仍然是边缘检测中的基本挑战之一，既影响传统方法也影响基于学习的方法。为获得清晰边缘，现有大多数方法依赖两种手工设计的后处理算法——非极大值抑制（NMS）与基于骨架的细化，它们不可微，阻碍了端到端优化。此外，所有现有的清晰边缘检测方法仍依赖此类后处理才能取得令人满意的结果。为克服这一局限，我们提出 MatchED，一个仅增加 21K 参数、即插即用的基于匹配的监督模块，可附加到任意边缘检测模型上，以端到端联合学习清晰边缘。在每次训练迭代中，MatchED 依据空间距离与置信度在预测边缘与真值边缘之间执行一对一匹配，从而保证训练与测试协议之间的一致性。在四个常用数据集上的大量实验表明，集成 MatchED 能显著提升现有边缘检测模型的性能。特别地，与基线模型相比，MatchED 将平均清晰度（AC）指标提升至多 2–4 倍。在强调清晰度的评测（CEval）下，MatchED 进一步将基线的 ODS 提升至多 20–35%，并在 OIS 与 AP 上取得类似增益，首次达到与标准后处理相当或超越的 SOTA 性能。代码见 https://cvpr26-matched.github.io。
英文原文摘要 1546 字符
Generating crisp, i.e., one-pixel-wide, edge maps remains one of the fundamental challenges in edge detection, affecting both traditional and learning-based methods. To obtain crisp edges, most existing approaches rely on two hand-crafted post-processing algorithms, Non-Maximum Suppression (NMS) and skeleton-based thinning, which are non-differentiable and hinder end-to-end optimization. Moreover, all existing crisp edge detection methods still depend on such post-processing to achieve satisfactory results. To address this limitation, we propose MatchED, a lightweight, only 21K additional parameters, and plug-and-play matching-based supervision module that can be appended to any edge detection model for joint end-to-end learning of crisp edges. At each training iteration, MatchED performs one-to-one matching between predicted and ground-truth edges based on spatial distance and confidence, ensuring consistency between training and testing protocols. Extensive experiments on four popular datasets demonstrate that integrating MatchED substantially improves the performance of existing edge detection models. In particular, MatchED increases the Average Crispness (AC) metric by up to 2--4x compared to baseline models. Under the crispness-emphasized evaluation (CEval), MatchED further boosts baseline performance by up to 20--35% in ODS and achieves similar gains in OIS and AP, achieving SOTA performance that matches or surpasses standard post-processing for the first time. Code is available at https://cvpr26-matched.github.io.