Research map / Point clouds / Point cloud classification
Minimal Adversarial Examples for Deep Learning on 3D Point Clouds
Jaeyeon Kim, Binh-Son Hua, Thanh Nguyen, Sai-Kit Yeung
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 random near-binary initialisation with critical-point initialisation (ablation) | minimal point perturbation attack · ScanObjectNN OBJ_BG | attack success rate: 90.16 -> 88.04 | Hurt |
| replace random initialisation with critical-point initialisation (ablation) | minimal point perturbation attack · ScanObjectNN OBJ_BG | Chamfer distance: 1.09 x 10^-4 -> 9.53 x 10^-4 | Hurt |
| replace random initialisation with critical-point initialisation (ablation) | minimal point perturbation attack · ScanObjectNN OBJ_BG | number of points: 37 -> 29 | Improved |
| initialise with all points instead of random points (ablation) | minimal point perturbation attack · ScanObjectNN OBJ_BG | number of points: 37 -> 60 | Hurt |
| initialise with all points instead of random points (ablation) | minimal point perturbation attack · ScanObjectNN OBJ_BG | attack success rate: 90.16 -> 91.12 | Improved |
| use the proposed l0-style minimal-point formulation instead of Xiang et al. perturbation | PointNet · ModelNet40 | attack success rate and points: 85.9 -> 89.38 success, 967 -> 36 points | Improved |
| use the proposed formulation instead of Xiang et al. point addition | PointNet · ModelNet40 | attack success rate: 73.59 -> 89.01 | Improved |
| use the proposed formulation instead of Xiang et al. perturbation | PointNet · ScanObjectNN OBJ_BG | attack success rate: 81.32 -> 91.72 | Improved |
| use the proposed formulation instead of random point selection | PointNet · ModelNet40 point perturbation | attack success rate: 55.56 -> 89.38 | Improved |
| attack DGCNN instead of PointNet | proposed attack · ModelNet40 | attack success rate: 89.38 -> 62.16 | Hurt |
| attack PointASNL instead of PointNet | proposed attack · ScanObjectNN OBJ_BG | attack success rate: 91.72 -> 55.88 | Hurt |
| attack PointNet++ instead of PointNet | proposed attack · ScanObjectNN OBJ_BG | attack success rate: 91.72 -> 94.05 | Improved |
Closest papers by meaning
- Benchmarking and Analyzing Robust Point Cloud Recognition: Bag of Tricks for Defending Adversarial Examples (ICCV 2023 · similarity 0.94)
- Generating 3D Adversarial Point Clouds (CVPR 2019 · similarity 0.93)
- Robust Adversarial Objects against Deep Learning Models (AAAI 2020 · similarity 0.93)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.89)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (CVPR 2025 · similarity 0.88)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 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.
