Research map / Adversarial robustness and security / Backdoor and poisoning
Towards Stealthy and Effective Backdoor Attacks on Lane Detection: A Naturalistic Data Poisoning Approach
Yifan Liao, Yuxin Cao, Yedi Zhang, Wentao He, Yan Xiao, Xianglong Du, Zhiyong Huang, Jin Song Dong
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 gradient-heatmap guided trigger placement (ablation) | DBALD (LaneATT) · CULane LDA | ASR: 80.97 -> 81.65 | No clear effect |
| add gradient-heatmap guided trigger placement (ablation) | DBALD (LaneATT) · CULane LRA | ASR: 70.16 -> 73.45 | Improved |
| add gradient-heatmap guided trigger placement (ablation) | DBALD (LaneATT) · CULane LOA | ASR: 68.52 -> 74.19 | Improved |
| add gradient-heatmap guided trigger placement (ablation) | DBALD (RESA) · CULane LRA | ASR: 58.26 -> 65.15 | Improved |
| add gradient-heatmap guided trigger placement (ablation) | DBALD (RESA) · CULane LOA | ASR: 69.37 -> 76.86 | Improved |
| add lane consistency loss to diffusion trigger generation (ablation) | DBALD · CULane forensic detection | UniDetection score (lower = stealthier): 12.13 -> 2.46 | Improved |
| add environment consistency loss to diffusion trigger generation (ablation) | DBALD · CULane forensic detection | UniDetection score (lower = stealthier): 14.51 -> 2.46 | Improved |
| use diffusion naturalistic trigger instead of BadLane texture | DBALD · CULane avg ASR, LaneATT | ASR: 54.56 -> 76.41 | Improved |
| use DBALD vs LD-Attack on ADNet | DBALD · TuSimple avg ASR, ADNet | ASR: 59.78 -> 80.66 | Improved |
| use DBALD vs LD-Attack on SCNN | DBALD · TuSimple avg ASR, SCNN | ASR: 92.76 -> 93.54 | No clear effect |
| apply rain, blur or occlusion to physical poisoned images | DBALD (LaneATT, cones) · Physical LOA | ASR: 57% -> 45/48/49% | Hurt |
| fine-tune poisoned model on clean data for 50 epochs (defense) | DBALD (LaneATT, LOA) · CULane LOA | ASR: 74.19% -> 63.68% | Hurt |
Closest papers by meaning
- Adversarial Backdoor Attack by Naturalistic Data Poisoning on Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.86)
- Lane2Seq: Towards Unified Lane Detection via Sequence Generation (CVPR 2024 · similarity 0.85)
- Temporal Logic-Based Multi-Vehicle Backdoor Attacks against Offline RL Agents in End-to-end Autonomous Driving (NeurIPS 2025 · similarity 0.85)
- GEONet: Global Enhancement and Optimization Network for Lane Detection (AAAI 2025 · similarity 0.84)
- Selective Focus: Investigating Semantics Sensitivity in Post-training Quantization for Lane Detection (AAAI 2024 · similarity 0.84)
- Beware of Road Markings: A New Adversarial Patch Attack to Monocular Depth Estimation (NeurIPS 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.
