Research map / Point clouds / Segmentation and detection
Subspace Prototype Guidance for Mitigating Class Imbalance in Point Cloud Semantic Segmentation
Jiawei Han, Kaiqi Liu, Wei Li, Guangzhi Chen
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 subspace prototype guidance (SPG) to PTv2 training | PTv2 · S3DIS Area5 | mIoU: 71.6 -> 73.3 | Improved |
| Add SPG to PointNet++ | PointNet++ · S3DIS Area5 | mIoU: 53.5 -> 64.0 | Improved |
| Add SPG to PTv2 on Toronto-3D | PTv2 · Toronto-3D | mIoU: 81.6 -> 83.1 | Improved |
| Use focal loss instead of SPG for class imbalance | PTv1 · S3DIS Area5 | mIoU: 70.4 -> 69.2 | Hurt |
| Use weighted CE instead of SPG for class imbalance | PTv2 · S3DIS Area5 | mIoU: 71.6 -> 70.9 | Hurt |
| Group points by category into separate feature subspaces in auxiliary branch (ablation) | SPG (PTv2) · S3DIS Area5 | mIoU: 72.2 -> 73.3 | Improved |
| Add supervised contrastive loss L_con (ablation) | SPG (PTv2) · S3DIS Area5 | mIoU: 72.6 -> 73.3 | Improved |
| Add auxiliary-to-main prototype guidance loss L_l1' (ablation) | SPG (PTv2) · S3DIS Area5 | mIoU: 71.6 -> 73.3 | Improved |
| Set EMA smoothing alpha=1-1/t instead of 0.9 (ablation) | SPG (PTv2) · S3DIS Area5 | mIoU: 72.1 -> 73.3 | Improved |
| Add SPG to PTv1 for Table category S3DIS (Table class) | PTv1 · S3DIS Area5 per-class table | IoU table: table 89.1 -> 83.3; chair 82.4 -> 91.0 | Mixed |
Closest papers by meaning
- Generating Inverse Feature Space for Class Imbalance in Point Cloud Semantic Segmentation (TPAMI 2025 · similarity 0.90)
- Dual-level Adaptive Self-Labeling for Novel Class Discovery in Point Cloud Segmentation (ECCV 2024 · similarity 0.87)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.86)
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Distribution Guidance Network for Weakly Supervised Point Cloud Semantic Segmentation (NeurIPS 2024 · similarity 0.85)
- Generated and Pseudo Content guided Prototype Refinement for Few-shot Point Cloud Segmentation (NeurIPS 2024 · 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.
