Research map / Point clouds / Point cloud classification
Semantic Privacy Protection with Utility Preservation for 3D Point Clouds
Jinchang Zhang, Jiakai Lin, David Crandall, Guoyu Lu
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 1024 input points instead of 512 (ablation) | Ours with PointMLP classifier · ScanObjectNN | NCRR: 0.781 -> 0.820 | Improved |
| use 1024 input points instead of 512 (ablation) | Ours with PointMLP classifier · ScanObjectNN | OCLR: 0.103 -> 0.080 | Improved |
| use 1024 input points instead of 512 (ablation) | Ours with PointNet classifier · ModelNet40 | NCRR: 0.821 -> 0.850 | Improved |
| add noise of magnitude 0.005 to the input points (ablation) | Ours with PointNet classifier · ScanObjectNN | NCRR: 0.651 -> 0.642 | Hurt |
| use a stronger classifier (PointMLP instead of PointNet) | Ours · ScanObjectNN | OCLR: 0.172 -> 0.080 | Improved |
| use a stronger classifier (PointMLP instead of PointNet) | Ours · ScanObjectNN | NCRR: 0.651 -> 0.820 | Improved |
| adaptive attacker retrains classifier instead of fixed attacker | Ours with PointMLP classifier · ScanObjectNN | original-label rate: 0.080 -> 0.098 | Hurt |
| class-transfer encryption instead of optical chaotic encryption | PointMLP on ScanObjectNN · ScanObjectNN | accuracy trained in encrypted domain: 0.099 -> 0.840 | Improved |
| class-transfer encryption instead of hybrid key encryption | PointMLP on ScanObjectNN · ScanObjectNN | accuracy trained in encrypted domain: 0.349 -> 0.840 | Improved |
| class-transfer encryption instead of UMT (permuted-wise test set) | PointMLP on ScanObjectNN · ScanObjectNN | accuracy trained in encrypted domain: 0.329 -> 0.840 | Improved |
| class-transfer encryption instead of UMT (class-wise test set) | PointMLP on ScanObjectNN · ScanObjectNN | accuracy trained in encrypted domain: 0.939 -> 0.840 | Hurt |
| class-transfer encryption instead of UMT (class-wise test set) | PointMLP on ScanObjectNN · ScanObjectNN | accuracy of original-trained classifier on encrypted data: 0.493 -> 0.119 | Improved |
Closest papers by meaning
- Unlearnable 3D Point Clouds: Class-wise Transformation Is All You Need (NeurIPS 2024 · similarity 0.86)
- PointCSP: Cross-Sample Semantic Propagation and Stability Preservation in Self-Supervised Point Cloud Learning (CVPR 2026 · similarity 0.85)
- All Rivers Run to the Sea: Private Learning with Asymmetric Flows (CVPR 2024 · similarity 0.84)
- MIRACLE 3D: Memory-efficient Integrated Robust Approach for Continual Learning on 3D Point Clouds via Shape Model Construction (ICLR 2025 · similarity 0.84)
- ConcreTizer: Model Inversion Attack via Occupancy Classification and Dispersion Control for 3D Point Cloud Restoration (ICLR 2025 · similarity 0.83)
- Toward Availability Attacks in 3D Point Clouds (ICML 2024 · similarity 0.83)
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.
