Research map / Point clouds / Segmentation and detection
CascadeProto: Cascaded Cross-Modal Prototype Purification via Entropy-Aware Learning for Few-Shot 3D Point Cloud Segmentation
Changshuo Wang, Weijun Li, Fan Mo, Zhonghang Liu, Shuting He, Prayag Tiwari, Dimitrios Kanoulas
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 Learnable Modality Adapter (ablation) | VIP-Seg baseline · S3DIS 2-way 1-shot (text) | mIoU: 81.28 -> 82.49 (+1.21%) | Improved |
| add Entropy Gate (ablation) | VIP-Seg + LMA · S3DIS 2-way 1-shot (text) | mIoU: 82.49 -> 83.91 (+1.42%) | Improved |
| add cascade of T=4 EPPM modules (ablation) | VIP-Seg + LMA + Entropy Gate · S3DIS 2-way 1-shot (text) | mIoU: 83.91 -> 85.97 (+2.06%) | Improved |
| add ADRM dynamic routing (ablation) | CascadeProto without ADRM · S3DIS 2-way 1-shot (text) | mIoU: 85.97 -> 86.53 (+0.56%) | Improved |
| increase cascade depth from T=1 to T=4 (ablation) | CascadeProto · S3DIS 2-way 1-shot (text) | mIoU: 83.28 -> 86.53 | Improved |
| increase cascade depth from T=4 to T=6 (ablation) | CascadeProto · S3DIS 2-way 1-shot (text) | mIoU: 86.53 -> 86.34 | No clear effect |
| use text modality instead of VIP-Seg alone | CascadeProto (Text) · S3DIS 2-way 1-shot | mIoU: +12.38% over VIP-Seg (74.15 -> 86.53) | Improved |
| use CascadeProto (Text) | VIP-Seg · ScanNet 2-way 1-shot | mIoU: 72.12 -> 79.57 (+7.45%) | Improved |
| swap text modality for audio | CascadeProto · S3DIS 3-way 5-shot | mIoU: 78.95 (text) -> 80.48 (audio) | Improved |
| add CascadeProto modules to VIP-Seg backbone | VIP-Seg · S3DIS S0 2-way 1-shot | mIoU: 72.20 -> 88.53, +0.12M params, +0.38 GFLOPs | Improved |
Closest papers by meaning
- Reasoning Beyond Points: A Visual Introspective Approach for Few-Shot 3D Segmentation (NeurIPS 2025 · similarity 0.88)
- ProtoTransfer: Cross-Modal Prototype Transfer for Point Cloud Segmentation (ICCV 2023 · similarity 0.88)
- From Coarse to Fine: Deep Prototype Refinement Network for Few-Shot Point Cloud Semantic Segmentation (ICML 2026 · similarity 0.87)
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.86)
- Generated and Pseudo Content guided Prototype Refinement for Few-shot Point Cloud Segmentation (NeurIPS 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.
