Research map / Point clouds / Point cloud classification
A Backdoor Attack Against 3D Point Cloud Classifiers
Zhen Xiang, David J. Miller, Siheng Chen, Xi Li, George Kesidis
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 spatial location optimization for backdoor points (ablation) | Backdoor attack P1 GS · ModelNet40 PointNet | ASR (%): 91.0 -> 94.0 | Improved |
| use backdoor attack instead of test-time evasion attack transferred from surrogate | PC backdoor vs TTE · ModelNet40 PointNet P5 | ASR (%): 44.7 -> 96.0 | Improved |
| use RS geometry instead of GS under anomaly detector (ablation) | Backdoor attack P2 · ModelNet40 PointNet with PC AD | ASR (%): 9.0 -> 97.0 | Improved |
| use RP geometry instead of GS under anomaly detector (ablation) | Backdoor attack P3 · ModelNet40 PointNet with PC AD | ASR (%): 51.0 -> 100 | Improved |
| deploy point anomaly detector against RS/RP/HS geometries (ablation) | Backdoor attack · ModelNet40 PointNet | ASR (%): minimum 81.0 | No clear effect |
| deploy point anomaly detector against GS geometry (ablation) | Backdoor attack · ModelNet40 PointNet P5 | ASR (%): 95.0 -> 2.0 | Hurt |
| change local geometry among GS, RS, RP, HS (with optimal location) (ablation) | Backdoor attack · ModelNet40 | ASR (%): 87% minimum ASR for all | No clear effect |
| use only 8 backdoor training samples (ablation) | Backdoor attack RP P1-P3 · ModelNet40 | ASR (%): ASR > 80% | Improved |
| reduce inserted backdoor points to 5 (ablation) | Backdoor attack RP P1-P3 · ModelNet40 | ASR (%): roughly 0.03125 chance all removed | Hurt |
| reduce inserted backdoor points to 5 under anomaly detector (ablation) | Backdoor attack RP P1-P3 · ModelNet40 with PC AD | ASR (%): roughly 0.5 probability | Hurt |
| increase epsilon from 0.005 to 0.2 (ablation) | Backdoor attack P1 RP · ModelNet40 | ASR (%): ASR >= 89% | No clear effect |
Closest papers by meaning
- Invisible Backdoor Attack against 3D Point Cloud Classifier in Graph Spectral Domain (AAAI 2024 · similarity 0.88)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.88)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (CVPR 2025 · similarity 0.87)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.87)
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.86)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 2024 · similarity 0.86)
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.
