遗忘而不丢失：从大规模视觉语言开放词表检测器中安全移除目标概念
Unlearning without Forgetting: Securely Removing Targeted Concepts from Large-Scale Vision-Language Open-Vocabulary Detectors
对做开放词表检测的视觉研究者，这篇给出了在冻结保留概念的前提下定向抹除指定概念-短语对的参数更新方案：离线构造保留知识嵌入的零空间，把遗忘更新投影到其正交补，并用一步 mean-flow 目标把遗忘概念推向不可检测；可直接复用于模型发布前的版权/隐私合规清理，以及检测器概念级可控编辑。
原文依据 constrains parameter updates to this orthogonal complement, mathematically preventing interference with retained concepts
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论开放词表检测器（OvOD）从网络规模预训练中继承了紧密耦合的跨模态知识，从而带来隐私、版权与合规风险。现有的机器遗忘方法在 OvOD 中面临几何纠缠干扰：由于可分解嵌入中共享的语义因子，遗忘更新不可避免地会扭曲被保留的知识。我们提出 SafeDetect，一个几何约束的遗忘框架，它离线地从保留知识嵌入中构造零空间，随后将参数更新约束到该正交补空间，从数学上防止对保留概念的干扰。遗忘通过一步均值流（mean-flow）目标实现，该目标将待遗忘概念推向不可检测状态，同时多模态解耦可防止跨模态恢复。我们建立了 UOD-Bench，这是首个面向 OvOD 遗忘的统一基准，包含 14.7K 张图像、67.3K 个区域-短语对，覆盖三项任务。在 UOD-Bench 与标准基准上、针对多种架构（如 GroundingDINO、LLM-Det）的大量实验表明，SafeDetect 取得了更优的遗忘效果（相较 NPO 提升 64.75%），同时保持稳定的保留性能与显著更好的零样本泛化能力，并且收敛速度比迭代式方法快 1.5 倍。
英文原文摘要 1326 字符
Open-vocabulary detectors (OvOD) inherit tightly coupled cross-modal knowledge from web-scale pretraining, creating privacy, copyright, and compliance risks. Existing machine unlearning methods face geometric entanglement interference in OvOD: forgetting updates inevitably distort preserved knowledge due to shared semantic factors in decomposable embeddings. We introduce SafeDetect, a geometrically constrained unlearning framework that constructs a null-space from preserved knowledge embeddings offline, then constrains parameter updates to this orthogonal complement, mathematically preventing interference with retained concepts. Forgetting is achieved through a one-step mean-flow objective that drives forgotten concepts toward non-detectable, while multimodal decoupling prevents cross-modal recovery. We establish UOD-Bench, the first unified benchmark for OvOD unlearning, featuring 14.7K images with 67.3K region-phrase pairs across three tasks. Extensive experiments across UOD-Bench and standard benchmarks with diverse architectures (e.g., GroundingDINO, LLM-Det) demonstrate that SafeDetect achieves superior forgetting efficacy (64.75% improvement over NPO) while maintaining stable retention performance and significantly better zero-shot generalization, with 1.5x faster convergence than iterative methods.