Research map / Adversarial robustness and security / Attacks on text and graphs
Latency NMS Attacks: Is It Real Life or Is It Just Fantasy?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Examines whether latency attacks targeting the NMS block of 2D object detectors, which work in lab settings, hold up in realistic deployment conditions.
Problem
Latency attacks against computer vision systems, especially adversarial images targeting Non-Max-Suppression in 2D object detection, may work on paper but fail in practice; the question is whether they are effective in realistic settings.
Approach
Tests previously reported NMS latency attacks under realistic conditions and finds that they do not hold up as reported, challenging the practical threat of such attacks.
Surprising findings
- Latency attacks on 2D object detection NMS, which appear effective on paper, do not necessarily work when tested in realistic settings.
Experimental setup
Metrics: latency
Builds on: NMS latency attacks
Closest papers by meaning
- Overload: Latency Attacks on Object Detection for Edge Devices (CVPR 2024 · similarity 0.89)
- Can't Slow Me Down: Learning Robust and Hardware-Adaptive Object Detectors against Latency Attacks for Edge Devices (CVPR 2025 · similarity 0.87)
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.86)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.84)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.83)
- Backdoor Attacks Against No-Reference Image Quality Assessment Models via a Scalable Trigger (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.