Phantom：以物理对象交互作为动态触发器劫持 NMS 的后门攻击
Phantom: Physical Object Interactions as Dynamic Triggers for NMS-Exploited Backdoors
把目标检测后处理中的 NMS 环节当作攻击面：用真实场景中的物体交互关系（而非固定 patch）作为动态触发器，可同时诱发误分类、误定位与目标凭空出现/消失，为检测模型的后门防御与 NMS 设计提供了新的威胁模型。
原文依据 By hijacking the Non-Maximum Suppression (NMS) process in object detectors
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论后门攻击对目标检测模型构成潜在威胁，凸显了研究其安全性的重要性。然而，现有后门攻击主要依赖触发器特有的内在特征，这限制了其在真实场景中的实用性。本文提出一种新型后门攻击，利用现实场景中的动态对象交互来激活恶意行为。该攻击通过劫持目标检测器中的非极大值抑制（NMS）过程，展现出稳健的有效性，包括误分类、误定位以及目标出现/消失，同时在干净输入上保持模型的正常性能。实验结果表明，该攻击在多种目标检测器和数据集上均表现出显著的攻击性能，并且在物理环境以及现有防御机制下仍然有效。这些发现凸显了开发高效且稳健的后门防御策略的迫切需求。
英文原文摘要 1031 字符
Backdoor attacks pose potential threats to object detection models, highlighting the importance of studying their security. However, existing backdoor attacks mainly rely on trigger-specific intrinsic features, which limits their practicality in real-world scenarios. In this paper, we propose a novel backdoor attack that leverages dynamic object interactions in realistic scenarios to activate malicious behavior. By hijacking the Non-Maximum Suppression (NMS) process in object detectors, this attack demonstrates robust effectiveness, including misclassification, mislocalization, and object appearance/disappearance, while maintaining the model's normal performance on clean inputs. Experimental results demonstrate that our attack exhibits significant attack performance across various object detectors and datasets, and remains effective both in physical environments and under existing defense mechanisms. These findings highlight the urgent need to develop efficient and robust defense strategies against backdoor attacks.