Research map / Point clouds / Point cloud classification
A Critical Revisit of Adversarial Robustness in 3D Point Cloud Recognition with Diffusion-Driven Purification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 supervised contrastive loss (LSupCon) to diffusion purification (ablation) | PointDP · ModelNet40 latent-feature adaptive attacks | robust accuracy: ~1.2% | Helped |
| add supervised contrastive loss (LSupCon) (ablation) | PointDP · ModelNet40 l_inf PGD at eps 0.075/0.1/0.125 | robust accuracy: gains grow with budget, e.g. +4.6 at eps 0.125 on PCT (40.5 to 45.0) | Helped |
| adversarial training on 3D point cloud models (kNN/FPS gradient obfuscation) | Adversarial training baseline · ModelNet40 l_inf eps=0.05 | robust accuracy under black-box SPSA vs white-box PGD: DGCNN 67.4 (PGD) vs 7.8 (SPSA) | Hurt |
| replace IF-Defense with PointDP diffusion purification | Purification defense · ModelNet40 average over six models | robust accuracy: +12.6% over IF-Defense | Helped |
| replace LPC with PointDP diffusion purification | Purification defense · ModelNet40 average | robust accuracy: +40.3% over LPC | Helped |
| diffusion purification (PointDP) instead of DUP-Net-style pipeline | Purification overhead · ModelNet40 inference time | time (s) per batch: 0.097 s vs 2.60 s (IF-Defense), ~27x speed-up | Helped |
| use PointDP instead of IF-Defense on ScanObjectNN | Purification defense · ScanObjectNN l_inf PGD | robust accuracy: 69.2 (PointDP) vs 3.1 (IF-Defense) on PointMLP | Helped |
| diffusion purification step count (t*) increased from small values (ablation) | PointDP · ModelNet40 point shift/add/drop PGD averaged | robust accuracy: converges after n>=30 (t=0.15) | Mixed |
Closest papers by meaning
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.89)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 2024 · similarity 0.89)
- Improving Robustness of 3D Point Cloud Recognition from a Fourier Perspective (NeurIPS 2024 · similarity 0.88)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.88)
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.88)
- Target-Guided Adversarial Point Cloud Transformer Towards Recognition Against Real-world Corruptions (NeurIPS 2024 · similarity 0.88)
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.