Research map / Point clouds / Point cloud classification
PointBA: Towards Backdoor Attacks in 3D Point Cloud
Xinke Li, Zhirui Chen, Yue Zhao, Zekun Tong, Yabang Zhao, Andrew Lim, Joey Tianyi Zhou
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 random rotation to target-class samples before implanting the trigger (ablation) | PointCBA (PointNet++) · ModelNet10 | ASR (%): 23.8 -> 40.2 | Improved |
| use BO-optimized rotation instead of random rotation for feature disentanglement (ablation) | PointCBA (PointNet++) · ModelNet10 | ASR (%): 40.2 -> 53.8 | Improved |
| use BO-optimized rotation instead of random rotation for feature disentanglement (ablation) | PointCBA (PointNet++) · ShapeNetPart | ASR (%): 41.0 -> 48.6 | Improved |
| add BO-optimized rotation to the clean-label trigger (versus no rotation) (ablation) | PointCBA (PointNet++) · ShapeNetPart | ASR (%): 19.0 -> 48.6 | Improved |
| increase orientation trigger angle from 1.25 to 2.5 degrees | PointPBA-O · ModelNet10 (PointNet++) | ASR (%): 27.9 -> 86.9 | Improved |
| increase orientation trigger angle from 5 to 40 degrees | PointPBA-O · ModelNet10 (PointNet++) | ASR (%): 92.1 -> 93.2 | No clear effect |
| increase injection rate up to 0.07 | PointCBA (PointNet++) · ModelNet10 | ASR (%): up to ASR 54% | Improved |
| vary injection rate | PointPBA-I and PointPBA-O (PointNet++) · ModelNet10 | ASR (%): over 90% | No clear effect |
| increase random shift factor lambda_beta of the interaction trigger | PointCBA (PointNet++) · ModelNet10 | ASR (%): down to 19.9% | Hurt |
| add random scale to the interaction trigger | PointCBA (PointNet++) · ModelNet10 | ASR (%) | No clear effect |
| add random jitter or scaling data augmentation during training | PointPBA and PointCBA · ModelNet10 | ASR (%): within 5% | Hurt |
| add rotation data augmentation during training | PointCBA and PointPBA-I · ModelNet10 | ASR (%): less than 10% | Hurt |
Closest papers by meaning
- Invisible Backdoor Attack against 3D Point Cloud Classifier in Graph Spectral Domain (AAAI 2024 · similarity 0.89)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (CVPR 2025 · similarity 0.89)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.88)
- Toward Availability Attacks in 3D Point Clouds (ICML 2024 · similarity 0.88)
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.88)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · 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.
