Research map / Detection and segmentation / CNN features
Self-Distillation for Robust LiDAR Semantic Segmentation in Autonomous Driving
Jiale Li, Hang Dai, Yong Ding
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 |
|---|---|---|---|
| replace average VFE with TransVFE (ablation) | U-Net baseline · SemanticKITTI val | mIoU: 64.90 -> 65.47 | Improved |
| replace average VFE with TransVFE (ablation) | U-Net baseline · nuScenes val | mIoU: 75.94 -> 76.42 | Improved |
| replace PointNet VFE with TransVFE (ablation) | U-Net baseline · SemanticKITTI val | mIoU: 65.08 -> 65.47 | Improved |
| add compound-transformation TTA instead of no TTA (ablation) | TransVFE baseline · SemanticKITTI val | mIoU: 65.47 -> 67.97 | Improved |
| use compound transform instead of four individual transformations in TTA (ablation) | TTA, M=5 · SemanticKITTI val | mIoU: 66.64 -> 67.75 | Improved |
| use compound transform instead of four individual transformations in TTA (ablation) | TTA, M=5 · nuScenes val | mIoU: 77.27 -> 77.54 | Improved |
| increase TTA variants beyond M=6 (ablation) | compound TTA · SemanticKITTI val | mIoU: 67.97 -> 68.04 | No clear effect |
| add TTA to teacher soft labels with EMA teacher (ablation) | self-distillation · SemanticKITTI val | mIoU: 66.18 -> 66.64 | Improved |
| add TTA to teacher soft labels with fixed pre-trained teacher (ablation) | self-distillation · SemanticKITTI val | mIoU: 65.89 -> 66.09 | Improved |
| update teacher by EMA of student instead of fixed pre-trained teacher (ablation) | self-distillation with TTA · SemanticKITTI val | mIoU: 66.09 -> 66.64 | Improved |
| initialize student and teacher from pre-trained baseline (ablation) | self-distillation EMA+TTA · SemanticKITTI val | mIoU: 66.64 -> 67.11 | Improved |
| add self-distillation (Exp. H) instead of baseline (ablation) | TransVFE baseline · nuScenes val | mIoU: 76.42 -> 77.74 | Improved |
Closest papers by meaning
- Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data (CVPR 2022 · similarity 0.91)
- Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation (CVPR 2022 · similarity 0.90)
- Temporal Consistent 3D LiDAR Representation Learning for Semantic Perception in Autonomous Driving (CVPR 2023 · similarity 0.90)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.88)
- Three Pillars Improving Vision Foundation Model Distillation for Lidar (CVPR 2024 · similarity 0.88)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.88)
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.
