Research map / Adversarial robustness and security / Attacks on text and graphs
Pretend Benign: A Stealthy Adversarial Attack by Exploiting Vulnerabilities in Cooperative Perception
Hongwei Lin, Dongyu Pan, Qiming Xia, Hai Wu, Cheng Wang, Siqi Shen, Chenglu Wen
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 |
|---|---|---|---|
| use region-selective attack (ARS + MAPG) instead of indiscriminate perturbation | cooperative perception attack · OPV2V, FLD defense | attack success rate bypassing defense: 0.0 -> 1.0 (vs PGD, C&W, James' Attack) | Improved |
| select attack regions with ARS instead of random 1.3% regions (ablation) | PGD, C&W, James' Attack · OPV2V, FLD defense | attack success rate: random regions 0.0 -> PB with ARS 1.0 | Improved |
| use feature-level defense FLD instead of ROBOSAC | defense against James' Attack · OPV2V validation | inference time and collaborator recall: 290.19 ms -> 30.17 ms; recall 0.37 -> 1.0 | Improved |
| use PB against FLD instead of ROBOSAC on V2XSet | PB · V2XSet | attack success rate: 0.98 -> 0.95 | No clear effect |
| increase time interval between historical and current frames in real-time PB (1 to 3) (ablation) | real-time PB · OPV2V, FLD defense | attack success rate: 0.99 -> 0.93 | Hurt |
| vary hyper-parameters mu_h and mu_l of region selection (ablation) | PB · OPV2V | attack success rate bypassing defense | No clear effect |
| add adversarial training with 50% PB-perturbed data | cooperative perception detectors · OPV2V, AP@IoU0.3 under PB | AP: 0.52 -> 0.88 (PointPillars+Average) | Improved |
| attack with an attention-fusion model instead of average fusion | PB · OPV2V no defense | AP@IoU0.3: 0.52 (Average) vs 0.58 (Attention) | Mixed |
Closest papers by meaning
- The Latent Guardian: Defending Collaborative Perception via Feature-Level Consistency Verification (ICML 2026 · similarity 0.86)
- All Vehicles Can Lie: Efficient Adversarial Defense in Fully Untrusted-Vehicle Collaborative Perception via Pseudo-Random Bayesian Inference (CVPR 2026 · similarity 0.86)
- Learning Mutual View Information Graph for Adaptive Adversarial Collaborative Perception (CVPR 2026 · similarity 0.86)
- AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (NeurIPS 2024 · similarity 0.84)
- Diffusion Guided Adversarial State Perturbations in Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.84)
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.
