Research map / Adversarial robustness and security / Attacks on text and graphs
Shadows Can Be Dangerous: Stealthy and Effective Physical-World Adversarial Attack by Natural Phenomenon
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Cast a shadow perturbation on the traffic sign | LISA-CNN and GTSRB-CNN · LISA and GTSRB test sets | attack success rate: 98.23% on LISA; 90.47% on GTSRB | Helped |
| Use a stronger shadow (lower k) (ablation) | Shadow Attack · LISA and GTSRB test sets | attack success rate: LISA-CNN: 72.71 -> 100.00; GTSRB-CNN: 66.73 -> 97.37 | Helped |
| Use more polygon edges (ablation) | Shadow Attack · LISA-CNN test set | attack success rate: 97.95 -> 98.91 -> 99.45 -> 99.59 | Helped |
| Use more polygon edges (ablation) | Shadow Attack · GTSRB-CNN test set | attack success rate: 90.80 -> 93.72 -> 96.83 -> 97.59 | Helped |
| Increase random restarts from 1 to 5 (ablation) | Shadow Attack · LISA-CNN test set | attack success rate: 95.91 -> 98.36 | Helped |
| Increase random restarts beyond 5 (ablation) | Shadow Attack · LISA and GTSRB test sets | attack success rate: LISA-CNN: 98.36 -> 99.18; GTSRB-CNN: 90.74 -> 93.22 | Helped |
| Train using random shadow perturbations | LISA-CNN · LISA test set | attack robustness: 1.73 -> 40.93 | Helped |
| Train using random shadow perturbations | GTSRB-CNN · GTSRB test set | attack robustness: 9.53 -> 25.57 | Helped |
| Train using random shadow perturbations | LISA-CNN · LISA test set | clean accuracy: 99.63 -> 99.56 | Hurt |
| Train using random shadow perturbations | GTSRB-CNN · GTSRB test set | clean accuracy: 99.00 -> 98.91 | Hurt |
Closest papers by meaning
- Invisible Perturbations: Physical Adversarial Examples Exploiting the Rolling Shutter Effect (CVPR 2021 · similarity 0.91)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.87)
- Full-Distance Evasion of Pedestrian Detectors in the Physical World (NeurIPS 2024 · similarity 0.86)
- ProjAttacker: A Configurable Physical Adversarial Attack for Face Recognition via Projector (CVPR 2025 · similarity 0.86)
- Infrared Adversarial Car Stickers (CVPR 2024 · similarity 0.85)
- The Fluorescent Veil: A Stealthy and Effective Physical Adversarial Patch Against Traffic Sign Recognition (NeurIPS 2025 · similarity 0.85)
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.