Research map / Point clouds / Segmentation and detection
Domain-aware Category-level Geometry Learning Segmentation for 3D Point Clouds
Pei He, Lingling Li, Licheng Jiao, Ronghua Shang, Fang Liu, Shuang Wang, Xu Liu, Wenping Ma
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 category-level geometry construction (CGC) (ablation) | baseline segmentation model · SemanticKITTI -> SemanticSTF | mIoU: 31.4 -> 32.8 | Improved |
| add geometry property learning loss L_GPL on top of CGC (ablation) | CGE · SemanticKITTI -> SemanticSTF | mIoU: 32.8 -> 33.0 | Improved |
| add physics-inspired adverse geometry simulation (PAGS) (ablation) | CGE model · SemanticKITTI -> SemanticSTF | mIoU: 33.0 -> 35.3 | Improved |
| add geometric consistency loss L_GCL aligning original and simulated embeddings (ablation) | CGE + PAGS · SemanticKITTI -> SemanticSTF | mIoU: 35.3 -> 37.5 | Improved |
| add test-time augmentation (rotate and scale) (ablation) | full DCGL · SemanticKITTI -> SemanticSTF | mIoU: 37.5 -> 39.1 | Improved |
| represent features in Wasserstein space instead of Euclidean space (ablation) | DCGL · SemanticKITTI -> SemanticSTF | mean mIoU: 35.2 -> 37.5 | Improved |
| add category-level embedding alignment (ablation) | DCGL · SemanticKITTI -> SemanticSTF | mean mIoU: 33.1 -> 37.5 | Improved |
| include PAGS simulation in training (ablation) | DCGL · SynLiDAR -> SemanticKITTI | mIoU: 30.1 -> 31.3 | Improved |
| DCGL versus best prior method (LiDARWeather) | MinkUNet18 · SemanticKITTI -> SemanticSTF | mean mIoU: 36.1 -> 37.5 | Improved |
| DCGL versus best prior method (DGUIL) on synthetic-to-real | MinkUNet18 · SynLiDAR -> SemanticKITTI | mIoU: 29.1 -> 31.3 | Improved |
| use MinkUNet34 backbone instead of MinkUNet18 | DCGL · SemanticKITTI -> SemanticSTF | mIoU: 37.6 -> 39.4 | Improved |
| raise momentum epsilon from 0.9 to 0.9999 for geometry matrix A (ablation) | DCGL · SemanticKITTI -> SemanticSTF | mIoU: 34.4 -> 39.4 | Improved |
Closest papers by meaning
- Geometry and Uncertainty-Aware 3D Point Cloud Class-Incremental Semantic Segmentation (CVPR 2023 · similarity 0.90)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.87)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 2024 · similarity 0.86)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.86)
- Collaborating Foundation Models for Domain Generalized Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration (AAAI 2025 · 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.
