Research map / Point clouds / Point cloud classification
DUP-Net: Denoiser and Upsampler Network for 3D Adversarial Point Clouds Defense
Hang Zhou, Kejiang Chen, Weiming Zhang, Han Fang, Wenbo Zhou, Nenghai Yu
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 |
|---|---|---|---|
| replace simple random sampling with statistical outlier removal (ablation) | point cloud defense · PointNet, C&W l2 attack | accuracy: 65.1% -> 81.4% | Improved |
| add PU-Net upsampler after SOR (DUP-Net vs SOR alone) (ablation) | SOR defense · PointNet, C&W l2 attack | accuracy: 81.4% -> 84.5% | Improved |
| add SOR before PU-Net upsampler (ablation) | PU-Net defense · PointNet, C&W l2 attack | accuracy: 23.9% -> 84.5% | Improved |
| add SOR before PU-Net upsampler (ablation) | PU-Net defense · PointNet, dropping 200 points | accuracy: 48.1% -> 43.7% | Hurt |
| add PU-Net upsampler (ablation) | no defense · PointNet, dropping 200 points | accuracy: 39.1% -> 48.1% | Improved |
| use SRS defense on dropping attack (ablation) | no defense · PointNet, dropping 200 points | accuracy: 39.1% -> 35.1% | Hurt |
| use DUP-Net instead of SRS on cluster/object attacks (ablation) | SRS defense · PointNet, C&W 3 clusters | accuracy: 92.0% -> 87.6% | Hurt |
| use DUP-Net instead of SRS on object-adding attack (ablation) | SRS defense · PointNet, C&W 3 objects | accuracy: 92.4% -> 68.4% | Hurt |
| add DUP-Net defense on clean inputs | PointNet · ModelNet40 clean | accuracy: 88.3% -> 86.3% | Hurt |
| replace PU-Net with simple Voronoi interpolation upsampling (ablation) | PU-Net defense · PointNet, dropping 200 points | accuracy: by 8% | Hurt |
| use Chamfer loss instead of EMD loss for PU-Net training (ablation) | PU-Net defense · PointNet, dropping attacks | accuracy | No clear effect |
| differentiable random point removal (p=0.5) before max pooling to simulate SOR (ablation) | modified PointNet · white-box attacks, C&W l2 | accuracy: 0% | No clear effect |
Later papers that compared with this method
2 later accepted papers report DUP-Net as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Certified L2-Norm Robustness of 3D Point Cloud Recognition in the Frequency Domain (AAAI 2026 · better on 6 of 6 shared results)
- Improving Adversarial Robustness of 3D Point Cloud Classification Models (ECCV 2022 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Benchmarking and Analyzing Robust Point Cloud Recognition: Bag of Tricks for Defending Adversarial Examples (ICCV 2023 · similarity 0.90)
- A Critical Revisit of Adversarial Robustness in 3D Point Cloud Recognition with Diffusion-Driven Purification (ICML 2023 · similarity 0.89)
- Improving Adversarial Robustness of 3D Point Cloud Classification Models (ECCV 2022 · similarity 0.89)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.88)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.88)
- You Should Learn to Stop Denoising on Point Clouds in Advance (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.
