Research map / Point clouds / Point cloud classification
Point Cloud Attacks in Graph Spectral Domain: When 3D Geometry Meets Graph Signal Processing
Daizong Liu, Wei Hu, Xin Li
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 GFT instead of voxelized DCT as spectral transform (ablation) | spectral attack on PointNet · PointNet under SOR 5% | attack success rate (%): 3.79 -> 86.75 | Improved |
| use GFT instead of voxelized DCT (ablation) | spectral attack on PointNet · PointNet no defense | attack success rate (%): 92.24 -> 100 | Improved |
| use GFT instead of DCT (ablation) | spectral attack on PointNet · PointNet | Chamfer distance D_c: 0.0038 -> 0.0006 | Improved |
| use GFT instead of DCT (ablation) | spectral attack on PointNet · PointNet | spectral energy E_delta: 32.4785 -> 0.7580 | Improved |
| vary K in the K-NN graph from 5 to 40 (ablation) | GSDA++ · PointNet | Hausdorff distance D_h: 0.0032 -> 0.0027 | No clear effect |
| vary frequency energy split from (75%,90%) to (73%,90%) (ablation) | GSDA++ · PointNet | Hausdorff distance D_h: 0.0028 -> 0.0030 | No clear effect |
| increase spectral constraint epsilon up to 1.5 (ablation) | GSDA++ · ModelNet40 PointNet | attack success rate: 100% at epsilon=1.5 | Improved |
| add GSDA++ improvements over GSDA under Gaussian noise 3% | GSDA · PointNet Gaussian noise defense | attack success rate (%): 72.16 -> 85.89 | Improved |
| add GSDA++ improvements over GSDA under SOR 20% | GSDA · PointNet SOR defense | attack success rate (%): 27.78 -> 29.31 | Improved |
| apply GSDA++ instead of GeoA under random scaling | GeoA · PointNet scaling | attack success rate (%): 39.73% -> 84.16% | Improved |
| apply GSDA++ instead of GeoA under random rotation | GeoA · PointNet rotation | attack success rate (%): 51.08% -> 87.69% | Improved |
| train victim with mixture of original and low-pass point clouds | PointNet vs GSDA++ attack · PointNet | attack success rate (%): 100% -> 12.58% | Improved |
Closest papers by meaning
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.90)
- Improving Robustness of 3D Point Cloud Recognition from a Fourier Perspective (NeurIPS 2024 · similarity 0.90)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.88)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 2024 · similarity 0.88)
- Manifold Constraints for Imperceptible Adversarial Attacks on Point Clouds (AAAI 2024 · similarity 0.87)
- Imperceptible 3D Point Cloud Attacks on Lattice-based Barycentric Coordinates (AAAI 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.
