Research map / Point clouds / Segmentation and detection
Less Is More: Reducing Task and Model Complexity for 3D Point Cloud Semantic Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use LiM3D rather than the strongest listed voxel baseline | SemanticKITTI semi-supervised segmentation · SemanticKITTI validation, 5% | mIoU (%): 50.6 -> 59.5 | Helped |
| Use LiM3D rather than the strongest listed voxel baseline | ScribbleKITTI semi-supervised segmentation · ScribbleKITTI validation, 5% | mIoU (%): 46.9$^*$ -> 58.1 | Helped |
| Add unreliable pseudo-labeling (ablation) | LiM3D · SemanticKITTI validation | mIoU (%): +0.7% on average | Helped |
| Add reflectivity features in training (ablation) | LiM3D with unreliable pseudo-labeling · SemanticKITTI validation | mIoU (%): +0.5% | Helped |
| Add Reflec-TTA without reflectivity features in training (ablation) | LiM3D with unreliable pseudo-labeling · SemanticKITTI validation | mIoU (%): +1.3% | Helped |
| Add both reflectivity features and Reflec-TTA (ablation) | LiM3D with unreliable pseudo-labeling · SemanticKITTI validation | mIoU (%): up to +2.8% | Helped |
| Use ST-RFD instead of uniform sampling (ablation) | LiM3D · SemanticKITTI validation | mIoU (%): +0.90%, +0.75%, +0.60% and +0.55% | Helped |
| Compute ST-RFD redundancy from range images rather than RGB images (ablation) | LiM3D with ST-RFD · SemanticKITTI and ScribbleKITTI validation | mIoU (%) | No clear effect |
| Select unreliable rather than reliable negative candidates (ablation) | LiM3D pseudo-labeling · SemanticKITTI validation, 5% | mIoU (%): 57.2 -> 59.5 | Helped |
| Select unreliable rather than random negative candidates (ablation) | LiM3D pseudo-labeling · SemanticKITTI validation, 5% | mIoU (%): 56.4 -> 59.5 | Helped |
| Use SDSC to reduce model parameters (ablation) | LiM3D · SemanticKITTI validation | mIoU (%): -1.4% | Hurt |
| Use reflectivity-based Reflec-TTA instead of intensity-based TTA (ablation) | LiM3D · SemanticKITTI and ScribbleKITTI validation | mIoU (%): 56.2, 59.1, 59.8, 60.9 -> 59.5, 62.2, 63.1, 63.3; 55.7, 57.5, 57.9, 59.2 -> 58.1, 61.0, 61.2, 62.0 | Helped |
Closest papers by meaning
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentation (CVPR 2021 · similarity 0.92)
- LaserMix for Semi-Supervised LiDAR Semantic Segmentation (CVPR 2023 · similarity 0.91)
- Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completion (AAAI 2021 · similarity 0.91)
- Collaborative Learning for Semi-Supervised LiDAR Semantic Segmentation (ICML 2026 · similarity 0.90)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.89)
- Beyond the Label Itself: Latent Labels Enhance Semi-supervised Point Cloud Panoptic Segmentation (AAAI 2024 · similarity 0.89)
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.