Research map / Point clouds / Segmentation and detection
Few-Shot Class-Incremental Learning for 3D Point Cloud Objects
Townim Chowdhury, Ali Cheraghian, Sameera Ramasinghe, Sahar Ahmadi, Morteza Saberi, Shafin Rahman
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 Microshape features (with language prototypes) (ablation) | FSCIL with PointNet · ShapeNet to CO3D FSCIL | final-task accuracy: 53.6 -> 57.1 | Improved |
| add Microshape features (with feature prototypes) (ablation) | FSCIL with PointNet · ShapeNet to CO3D FSCIL | final-task accuracy: 51.8 -> 53.4 | Improved |
| use language prototypes instead of feature prototypes (ablation) | FSCIL with Microshape · ShapeNet to CO3D FSCIL | final-task accuracy: 53.4 -> 57.1 | Improved |
| freeze feature extractor during incremental steps (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 50.7 -> 57.1 | Improved |
| add SVD orthogonalization of k-means centers (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 54.6 -> 57.1 | Improved |
| use cross-entropy loss instead of MSE in the relation network (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 55.2 -> 57.1 | Improved |
| increase k-means centroids from 256 to 1024 (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 44.8 -> 57.1 | Improved |
| increase k-means centroids from 1024 to 2048 (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 57.1 -> 57.3 | No clear effect |
| increase k-means centroids from 1024 to 4096 (ablation) | Microshape FSCIL · ShapeNet to CO3D FSCIL | final-task accuracy: 57.1 -> 57.0 | No clear effect |
| add Microshape features and language prototype (ablation) | dynamic few-shot learning, 1-shot · ModelNet to ScanObjectNN dynamic few-shot | accuracy: 39.90 -> 72.23 | Improved |
| add Microshape features and language prototype (ablation) | dynamic few-shot learning, 5-shot · ModelNet to ScanObjectNN dynamic few-shot | accuracy: 51.65 -> 72.53 | Improved |
| add Microshape-based backbone (ablation) | 3D recognition baseline · ModelNet40 to ScanObjectNN recognition (11 common classes) | accuracy: 44.25 -> 46.34 | Improved |
Closest papers by meaning
- Geometric Feature Embedding for Effective 3D Few-Shot Class Incremental Learning (ICML 2025 · similarity 0.91)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.89)
- Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · similarity 0.86)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.86)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.86)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 2024 · 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.
