Research map / Adversarial robustness and security / Attacks on text and graphs
Invisible Triggers, Visible Threats! Road-Style Adversarial Creation Attack for Visual 3D Detection in Autonomous Driving
Jian Wang, Lijun He, Yixing Yong, Haixia Bi, Fan Li
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 Scenario-Associated Adaptation (stage 2) on top of stage-1 generator (ablation) | AdvRoad · nuScenes val, BEVDet-R50 | ASR (CD 2.0m): 23.4 -> 62.6 | Improved |
| train generator with adversarial objective instead of a plain road generator (ablation) | AdvRoad · nuScenes val, BEVDet-R50 | ASR (CD 2.0m): 26.7 -> 62.6 | Improved |
| extend stage-2 update iterations to 50 (ablation) | AdvRoad · nuScenes val, BEVDet-R50 | ASR (CD 2.0m): 62.6 -> 67.0 | Improved |
| increase poster length from 3.0m to 5.0m (width 2.0m) (ablation) | AdvRoad · nuScenes val, BEVDet-R50 | ASR (CD 2.0m): 49.4 -> 67.6 (gains diminish beyond 4m; CD0.5m 23.3 -> 23.1) | Improved |
| use AdvRoad road-style poster instead of real vehicle picture poster | visual 3D detectors · nuScenes val, BEVFormer-R50 | ASR (CD 2.0m): 6.3 -> 44.5 | Improved |
| use AdvRoad instead of AdvPoster (pixel-optimized single poster), no defense | BEVDet and BEVFormer · nuScenes val | ASR (CD 2.0m): AdvPoster 91% and 82.6% (higher than AdvRoad) | Hurt |
| use AdvRoad instead of AdvPoster against adversarially fine-tuned detectors | defended detectors · nuScenes val | ASR (CD 2.0m): AdvPoster <2% vs AdvRoad ~20% | Improved |
| use AdvRoad instead of real picture poster for perceptual similarity | BEVDet-R50 · nuScenes val | LPIPS: 0.2066 -> 0.1472 | Improved |
| switch from sunlit to shaded or indoor physical environment | AdvRoad physical poster · physical attack | ASR: 49.4% -> 28.3% (shaded), 19.5% (indoor) | Hurt |
Closest papers by meaning
- SABER: Spatially Consistent 3D Universal Adversarial Objects for BEV Detectors (CVPR 2026 · similarity 0.90)
- Beware of Road Markings: A New Adversarial Patch Attack to Monocular Depth Estimation (NeurIPS 2024 · similarity 0.89)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (CVPR 2025 · similarity 0.88)
- Physical 3D Adversarial Attacks against Monocular Depth Estimation in Autonomous Driving (CVPR 2024 · similarity 0.87)
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.87)
- Infrared Adversarial Car Stickers (CVPR 2024 · 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.
