Research map / Point clouds / Point cloud classification
3D-ANC: Adaptive Neural Collapse for Robust 3D Point Cloud Recognition
Yuanmin Huang, Wenxuan Li, Mi Zhang, Xiaohan Zhang, Xiaoyu You, Min Yang
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 the classifier with a fixed simplex ETF head (ablation) | PointNet · ModelNet40 (avg over 5 attacks) | accuracy under attack: +38.1 | Improved |
| replace the classifier with a fixed simplex ETF head (ablation) | PointNet · ModelNet40 (clean) | clean accuracy: -0.6 | Hurt |
| add RBL (learnable rotation) to the ETF head (ablation) | 3D-ANC (ETF head) · ModelNet40 (clean) | clean accuracy: -0.6 -> +3.7 | Improved |
| add RBL (learnable rotation) to the ETF head (ablation) | 3D-ANC (ETF head) · ModelNet40 (avg over 5 attacks) | accuracy under attack: +38.1 -> +34.5 | Hurt |
| add FDL to the ETF head without RBL (ablation) | 3D-ANC (ETF head) · ModelNet40 (avg over 5 attacks) | accuracy under attack: +38.1 -> +31.7 | Hurt |
| add FDL to the ETF head without RBL (ablation) | 3D-ANC (ETF head) · ModelNet40 (clean) | clean accuracy: -0.6 -> -1.4 | Hurt |
| add FDL on top of ETF and RBL (ablation) | 3D-ANC (ETF & RBL) · ModelNet40 (avg over 5 attacks) | accuracy under attack: +34.5 -> +39.3 | Improved |
| add FDL on top of ETF and RBL (ablation) | 3D-ANC (ETF & RBL) · ModelNet40 (clean) | clean accuracy: +3.7 -> +0.9 | Hurt |
| apply 3D-ANC (ETF head, RBL, FDL) | DGCNN · ModelNet40 (avg over 5 attacks) | accuracy under attack: 27.2 -> 80.9 | Improved |
| apply 3D-ANC (ETF head, RBL, FDL) | PCT · ModelNet40 (avg over 5 attacks) | accuracy under attack: 47.5 -> 77.3 | Improved |
| apply 3D-ANC versus best baseline defense on shape-invariant attacks | PointNet · ModelNet40 (shape-invariant attacks, avg) | accuracy under attack: 54.9 -> 63.6 | Improved |
| apply 3D-ANC versus PointCutMix on HiT attack | PointNet · ModelNet40 (HiT attack) | accuracy: 73.7 -> 66.1 | Hurt |
Closest papers by meaning
- Benchmarking and Analyzing Robust Point Cloud Recognition: Bag of Tricks for Defending Adversarial Examples (ICCV 2023 · similarity 0.89)
- Robust Structured Declarative Classifiers for 3D Point Clouds: Defending Adversarial Attacks With Implicit Gradients (CVPR 2022 · similarity 0.89)
- Improving Adversarial Robustness of 3D Point Cloud Classification Models (ECCV 2022 · similarity 0.89)
- Shape-Invariant 3D Adversarial Point Clouds (CVPR 2022 · similarity 0.89)
- Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack (AAAI 2024 · similarity 0.87)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 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.
