Research map / Point clouds / Point cloud classification
Exploring the Devil in Graph Spectral Domain for 3D Point Cloud Attacks
Qianjiang Hu, Daizong Liu, Wei Hu
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 graph Fourier transform instead of 3D DCT as spectral representation (ablation) | GSDA · ModelNet40 PointNet SOR 10% | attack success rate (%): 25.17 -> 72.53 | Improved |
| use graph Fourier transform instead of 3D DCT as spectral representation (ablation) | GSDA · ModelNet40 PointNet | spectral perturbed energy E_delta: 19.4804 -> 3.4510 | Improved |
| vary K-NN graph size K from 5 to 40 (ablation) | GSDA · ModelNet40 PointNet | Chamfer distance D_c: 0.0007 -> 0.0007 | No clear effect |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet · ModelNet40 targeted attack | D_norm: 0.4385 -> 0.1741 | Improved |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet++ · ModelNet40 targeted attack | Hausdorff distance D_h: 0.0357 -> 0.0248 | Improved |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | DGCNN · ModelNet40 targeted attack | Hausdorff distance D_h: 0.0402 -> 0.1401 | Hurt |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet · ModelNet40 SOR defense 20% | attack success rate (%): 11.71 -> 27.78 | Improved |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet · ModelNet40 IF-Defense | attack success rate (%): 38.72% -> 50.26% | Improved |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet to PointNet++ transfer · ModelNet40 | transfer success rate (%): 11.59% -> 11.51% | No clear effect |
| perturb in spectral domain rather than data domain (GSDA vs GeoA) | PointNet++ to DGCNN transfer · ModelNet40 | transfer success rate (%): 12.46% -> 32.31% | Improved |
| relax spectral perturbation bound epsilon_max up to 3.0 (ablation) | GSDA · ModelNet40 | attack success rate: almost 100% | Improved |
| attack only low or high frequency band instead of the whole spectrum (ablation) | GSDA · ModelNet40 | imperceptibility (qualitative) | Hurt |
Closest papers by meaning
- Improving Robustness of 3D Point Cloud Recognition from a Fourier Perspective (NeurIPS 2024 · similarity 0.90)
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.89)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 2024 · similarity 0.87)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.87)
- 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 0.86)
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.
