Research map / Adversarial robustness and security / Attacks on text and graphs
A Comprehensive Study of the Robustness for LiDAR-Based 3D Object Detectors Against Adversarial Attacks
Yifan Zhang, Junhui Hou, Yixuan Yuan
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 re-voxelization of perturbed points to PGD attack (ablation) | PGD point perturbation attack · KITTI SECOND (eps=10cm) | mAP ratio (%): 98.63 -> 93.77 | Improved |
| add flipped perturbation direction on top of re-voxelization (ablation) | PGD point perturbation attack · KITTI SECOND (eps=10cm) | mAP ratio (%): 93.77 -> 70.80 | Improved |
| add flipped perturbation with large voxel size [0.20,0.20,0.4] (ablation) | PGD point perturbation attack · KITTI SECOND (eps=10cm) | mAP ratio (%): 89.55 -> 101.78 | Hurt |
| re-voxelize perturbed points with small eps, no flipping (ablation) | PGD point perturbation attack · KITTI SECOND (eps=7cm) | mAP ratio (%): 103.19 | Hurt |
| add D_cluster to pillar encoding (ablation) | PointPillar · KITTI PGD perturbation (eps=10cm) | mAP ratio (%): 64.25 -> 25.24 | Hurt |
| add D_center to pillar encoding (ablation) | PointPillar · KITTI PGD perturbation (eps=10cm) | mAP ratio (%): 64.25 -> 11.59 | Hurt |
| use full pillar encoding (both D_cluster and D_center) (ablation) | PointPillar · KITTI PGD perturbation (eps=10cm) | mAP ratio (%): 64.25 -> 7.28 | Hurt |
| change pillar size in x and y (ablation) | PointPillar · KITTI PGD perturbation (eps=10cm) | mAP ratio (%): 1.76 to 7.53 across sizes | No clear effect |
| use PGD instead of FGM for point perturbation (iterative vs single step) | point perturbation attack · KITTI average over 8 detectors | mAP ratio (%): 70.01 -> 47.29 | Improved |
| use cross-task attack from segmentation instead of source PGD attack | point perturbation attack · KITTI SECOND | mAP ratio (%): 70.80 -> 48.09 | Improved |
| use cross-task PGD attack instead of source PGD attack on PointPillar | point perturbation attack · KITTI PointPillar | mAP ratio (%): 7.28 -> 80.23 | Hurt |
| use PointRCNN as source model for transfer attack instead of SECOND self-attack | PGD perturbation · KITTI SECOND | mAP ratio (%): 70.80 -> 49.77 | Improved |
Closest papers by meaning
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.89)
- A New Adversarial Perspective for LiDAR-based 3D Object Detection (AAAI 2025 · similarity 0.89)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.87)
- NBA3D: Neighbor-Based Confidence Adjustment for 3D Rare Object Detection Using LiDAR (AAAI 2025 · similarity 0.87)
- Fusion Is Not Enough: Single Modal Attacks on Fusion Models for 3D Object Detection (ICLR 2024 · similarity 0.87)
- MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection (ICLR 2025 · similarity 0.87)
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.
