Research map / Point clouds / Segmentation and detection
White Aggregation and Restoration for Few-shot 3D Point Cloud Semantic Segmentation
Jiyun Im, SuBeen Lee, Miso Lee, Jae-Pil Heo
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 FPS prototypes with WARM | COSeg · S3DIS (1-way 1-shot, S0) | mIoU: 46.31 -> 59.26 | Improved |
| replace FPS prototypes with WARM | MM-FSS · S3DIS (1-way 1-shot, S0) | mIoU: 49.84 -> 52.62 | Improved |
| replace FPS + min-dist. prototypes with WARM | FPS + min-dist. · S3DIS (overall mean) | mIoU: 51.03 -> 55.66 | Improved |
| replace FPS + min-dist. prototypes with WARM | FPS + min-dist. · ScanNet (overall mean) | mIoU: 39.01 -> 41.21 | Improved |
| add whitening and coloring around cross-attention (ablation) | naive cross-attention · S3DIS (1-way 1-shot, S0) | mIoU | Improved |
| add coloring after whitening (ablation) | WARM · S3DIS (1-way 1-shot, S0) | mIoU | Improved |
| add simplification loss to margin loss (ablation) | WARM · S3DIS (1-way 1-shot, S0) | mIoU | Improved |
| use adaptive threshold lambda_median(1+sqrt(D/L))^2 instead of epsilon (ablation) | WARM · S3DIS (S0, L>D episodes) | mIoU: 60.37 -> 61.07 | Improved |
| use adaptive threshold instead of lambda_mean (ablation) | WARM · S3DIS (S0, L>D episodes) | mIoU: 57.46 -> 61.07 | Improved |
| use lambda_median*4 instead of the L>D threshold when L<=D (ablation) | WARM · S3DIS (S0, L<=D episodes) | mIoU: 59.01 -> 59.79 | Improved |
| increase backbone capacity of FPS + min-dist. from 3.91M to 5.16M parameters (ablation) | FPS + min-dist. · S3DIS | mIoU | No clear effect |
| reduce prototypical tokens to M=10 (ablation) | WARM · ScanNet | mIoU | Hurt |
Closest papers by meaning
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.92)
- Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · similarity 0.89)
- Multimodality Helps Few-shot 3D Point Cloud Semantic Segmentation (ICLR 2025 · similarity 0.89)
- Generated and Pseudo Content guided Prototype Refinement for Few-shot Point Cloud Segmentation (NeurIPS 2024 · similarity 0.88)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.88)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · 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.
