Research map / Point clouds / Segmentation and detection
Unsupervised Point Cloud Registration via Training-Time Semantic Guidance
Kezheng Xiong, Shiyun Xu, Sheng Ao, Siqi Shen, Cheng Wang, Chenglu Wen
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 Semantic-aware Teacher Refining (SATR) (ablation) | CAESAR · KITTI/nuScenes ablation | mRR: 1.7% mRR drop when removed | Improved |
| use L_rerank instead of global L_reg inside SATR (ablation) | CAESAR · KITTI/nuScenes ablation | mRR: -1.5% mRR when replaced by L_reg | Improved |
| use both s_geo and s_sem in DCRM scoring (ablation) | CAESAR · KITTI/nuScenes ablation | registration performance | Improved |
| add Reliable Label Selection (ablation) | CAESAR · KITTI/nuScenes ablation | tIR@1st epoch | Improved |
| add Semantic Predictive Distillation (ablation) | CAESAR · KITTI/nuScenes ablation | mRR | Improved |
| use SPD instead of vanilla L1 feature mimicry (ablation) | CAESAR · KITTI/nuScenes ablation | mRR | Improved |
| train with CAESAR instead of INTEGER (cross-dataset) | CAESAR · KITTI to nuScenes (cross-dataset) | mRR: +11.5% mRR vs INTEGER | Improved |
| train with CAESAR instead of INTEGER (long range, cross-dataset) | CAESAR · KITTI to nuScenes (cross-dataset) | RR@[40,50): +6.9% in RR@[40,50) | Improved |
| use SATR instead of full teacher adaptation | CAESAR · Teacher refinement cost | trainable parameters (M): 8.76 -> 0.07 | Improved |
| use unsupervised DGT-ST semantics instead of supervised segmenter (ablation) | CAESAR · KITTI/nuScenes | mRR | Mixed |
| add semantic anchoring (CAESAR vs INTEGER) | CAESAR · nuScenes (test) | mRR: 63.1 -> 79.5 | Improved |
| add semantic anchoring (CAESAR vs INTEGER) | CAESAR · KITTI (test) | mRR: 84.0 -> 86.5 | Improved |
Closest papers by meaning
- LESS: Label-Efficient Semantic Segmentation for LiDAR Point Clouds (ECCV 2022 · similarity 0.91)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.90)
- Complete & Label: A Domain Adaptation Approach to Semantic Segmentation of LiDAR Point Clouds (CVPR 2021 · similarity 0.90)
- Cross-Modal Prior-Guided Training with Visual Foundation Models for Unsupervised LiDAR Point Cloud Registration (NeurIPS 2026 · similarity 0.90)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.88)
- Construct to Associate: Cooperative Context Learning for Domain Adaptive Point Cloud Segmentation (CVPR 2024 · similarity 0.87)
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.
