Research map / Adversarial robustness and security / Attacks on text and graphs
New Threats against Object Detector with Non-local Block
Yi Huang, Fan Wang, Adams Wai-Kin Kong, Kwok-Yan Lam
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add a non-local block to Faster R-CNN backbone | Faster R-CNN ResNet-101 · Digital disappearing attack (stop sign) | successful disappearing attack rate (%): 7.6 -> 52.7 | Hurt |
| add a non-local block to Faster R-CNN backbone | Faster R-CNN ResNet-101 · Digital appearing attack (boat), IOD=0 | successful appearing attack rate (%): 0 -> 53.5 | Hurt |
| add a non-local block to Faster R-CNN backbone | Faster R-CNN ResNet-101 · Digital appearing attack (boat), IOD<0.4 | successful appearing attack rate (%): 6.6 -> 55.4 | Hurt |
| add a non-local block to Faster R-CNN backbone | Faster R-CNN ResNet-101 · Stop sign detection without attack | original detection rate (%): 78.6 -> 80.6 | Improved |
| apply the universal adversarial patch in the physical world against Faster R-CNN-WN (disappearing) | Faster R-CNN-WN · Physical disappearing attack (stop sign) | detection rate (%): 89.9 -> 47.6 | Hurt |
| apply the universal adversarial patch in the physical world against Faster R-CNN-WN (appearing) | Faster R-CNN-WN · Physical appearing attack (boat) | appearing rate (%): 0.8 -> 31.3 | Hurt |
| place the patch at locations different from the training position (11x11 grid) (ablation) | Faster R-CNN-WN disappearing patch · Effective attack regions, 670 images | successful attack rate: most rates higher than 30% | No clear effect |
| remove the non-local block when patch is placed at different locations (ablation) | Faster R-CNN disappearing patch · Effective attack regions, 645 images | successful attack rate | Mixed |
| set epsilon too small or too large in the appearing attack objective (ablation) | appearing attack patch training · Appearing attack, boat | attack strength (qualitative) | Hurt |
Closest papers by meaning
- Non-Local Neural Networks (CVPR 2018 · similarity 0.84)
- Beyond Digital Domain: Fooling Deep Learning Based Recognition System in Physical World (AAAI 2020 · similarity 0.84)
- Segment and Complete: Defending Object Detectors Against Adversarial Patch Attacks With Robust Patch Detection (CVPR 2022 · similarity 0.84)
- PBCAT: Patch-Based Composite Adversarial Training against Physically Realizable Attacks on Object Detection (ICCV 2025 · similarity 0.83)
- Transferable Adversarial Attacks for Object Detection Using Object-Aware Significant Feature Distortion (AAAI 2024 · similarity 0.83)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.82)
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.
