Research map / Point clouds / Point cloud classification
Canonical Shape Projection is All You Need for 3D Few-shot Class Incremental Learning
Ali Cheraghian, Zeeshan Hayder, Sameeea Ramasinghe, Shafin Rahman, Javad Jafaryahya, Lars Petersson, Mehrtash Harandi
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 |
|---|---|---|---|
| add the learned rotation (canonical projection) loss Lc on top of Lr (ablation) | C3PR · ShapeNet -> ScanObjectNN | relative accuracy drop Delta: 22.9 -> 17.3 | Improved |
| add gradient consistency loss Lg on top of Lr (ablation) | C3PR · ShapeNet -> ScanObjectNN | relative accuracy drop Delta: 22.9 -> 20.6 | Improved |
| add Lg on top of Lr + Lc (ablation) | C3PR · ShapeNet -> ScanObjectNN | relative accuracy drop Delta: 17.3 -> 14.9 | Improved |
| add module C background reprogramming (ablation) | C3PR · ShapeNet -> ScanObjectNN | relative accuracy drop Delta: 18.4 -> 14.9 | Improved |
| use consistent-background reprogramming instead of conventional reprogramming (ablation) | C3PR · ShapeNet -> ScanObjectNN | relative accuracy drop Delta: 17.3 -> 14.9 | Improved |
| replace ten fixed camera depth maps (PointCLIP v2) with one learned projection (ablation) | C3PR · ShapeNet -> ScanObjectNN | accuracy per task | Improved |
| use multiple GPT-generated prompts per class instead of one fixed prompt (ablation) | C3PR · ShapeNet -> ScanObjectNN | accuracy per task | Improved |
| apply learned projection and reprogramming to point cloud classification | CurveNet · ModelNet40 classification | accuracy: 93.8 -> 94.6 | Improved |
| train on synthetic base classes then few-shot real-scanned classes (domain gap) | C3PR · cross-dataset vs within-dataset | relative accuracy drop Delta | Hurt |
Closest papers by meaning
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.87)
- Geometric Feature Embedding for Effective 3D Few-Shot Class Incremental Learning (ICML 2025 · similarity 0.87)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.86)
- Duoduo CLIP: Efficient 3D Understanding with Multi-View Images (ICLR 2025 · similarity 0.86)
- UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025 · similarity 0.86)
- Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition (CVPR 2025 · similarity 0.85)
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.
