Research map / Point clouds / Segmentation and detection
LESS: Label-Efficient Semantic Segmentation for LiDAR Point Clouds
Minghua Liu, Yin Zhou, Charles R. Qi, Boqing Gong, Hao Su, Dragomir Anguelov
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 pre-segmentation with component-wise labeling (ablation) | random point labels, Cylinder3D · SemanticKITTI val (0.1% labels) | mIoU: 48.1 -> 59.3 | Improved |
| add weak labels derived from components (ablation) | pre-segmentation + sparse labels · SemanticKITTI val (0.1% labels) | mIoU: 59.3 -> 61.6 | Improved |
| add propagated labels from pure components (ablation) | pre-segmentation + sparse labels · SemanticKITTI val (0.1% labels) | mIoU: 59.3 -> 62.2 | Improved |
| combine weak and propagated labels (ablation) | pre-segmentation + sparse labels · SemanticKITTI val (0.1% labels) | mIoU: 59.3 -> 63.5 | Improved |
| add contrastive prototype learning (ablation) | LESS · SemanticKITTI val (0.1% labels) | mIoU: 63.5 -> 64.9 | Improved |
| add multi-scan distillation (ablation) | LESS · SemanticKITTI val (0.1% labels) | mIoU: 64.9 -> 66.0 | Improved |
| label 0.1% random points versus 0.1% random scans | Cylinder3D · SemanticKITTI val (pilot) | mIoU: 35.6 vs 48.0 (random scans vs random points) | Improved |
| use LESS pre-segmentation instead of uniform grid partition (ablation) | LESS · SemanticKITTI val (0.1% labels) | mIoU: 61.4 -> 64.9 | Improved |
| use LESS pre-segmentation instead of geometric partition (ablation) | LESS · SemanticKITTI val (0.1% labels) | mIoU: 61.9 -> 64.9 | Improved |
| use LESS pre-segmentation instead of active labeling (ablation) | LESS · SemanticKITTI val (0.1% labels) | mIoU: 54.2 -> 64.9 | Improved |
| add prototype learning and multi-scan distillation to fully supervised training | Cylinder3D · SemanticKITTI val (100% labels) | mIoU: 65.9 -> 66.2 | No clear effect |
| use multi-scan teacher versus single-scan model | LESS · SemanticKITTI val (0.1% labels) | mIoU: 64.9 -> 66.8 | Improved |
Closest papers by meaning
- Less Is More: Reducing Task and Model Complexity for 3D Point Cloud Semantic Segmentation (CVPR 2023 · similarity 0.94)
- Beyond the Label Itself: Latent Labels Enhance Semi-supervised Point Cloud Panoptic Segmentation (AAAI 2024 · similarity 0.90)
- Less Is More: Label Recommendation for Weakly Supervised Point Cloud Semantic Segmentation (AAAI 2024 · similarity 0.90)
- Exploring Scene Affinity for Semi-Supervised LiDAR Semantic Segmentation (CVPR 2025 · similarity 0.90)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.90)
- 3D-AVS: LiDAR-based 3D Auto-Vocabulary Segmentation (CVPR 2025 · similarity 0.90)
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.
