Research map / Point clouds / Segmentation and detection
Domain Generalization of 3D Semantic Segmentation in Autonomous Driving
Jules Sanchez, Jean-Emmanuel Deschaud, François Goulette
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 |
|---|---|---|---|
| remove reflectivity from the input of segmentation networks (ablation) | KPConv · SemanticKITTI -> SemanticPOSS | mIoU: 33.1 -> 39.1 | Improved |
| remove reflectivity from the input (ablation) | Cylinder3D · SemanticKITTI -> SemanticPOSS | mIoU: 33.8 -> 41.9 | Improved |
| remove reflectivity from the input (ablation) | SRUNet · SemanticKITTI -> SemanticPOSS | mIoU: 45.2 -> 45.3 | No clear effect |
| remove reflectivity from the input (ablation) | Cylinder3D · SemanticKITTI source-to-source | mIoU: 66.9 -> 60.7 | Hurt |
| accumulate past scans for KPConv (ablation) | KPConv · SemanticKITTI -> SemanticPOSS | mIoU: 39.1 -> 47.2 | Improved |
| naive multi-frame accumulation without propagation and clustering (ablation) | KPConv · SemanticKITTI source-to-source | mIoU: 58.3 -> 53.0 | Hurt |
| add label propagation and clusterization to multiframe KPConv (ablation) | KPConv multiframe · SemanticKITTI -> SemanticPOSS | mIoU / FPS: mIoU 47.2 -> 50.4, FPS 0.05 -> 1.0 | Improved |
| use 3DLabelProp instead of single-scan baselines on solid-state lidar | KPConv · SemanticKITTI -> PandaFF | mIoU: 22.2 (best baseline) -> 55.2 | Improved |
| use 3DLabelProp on nuScenes target | KPConv · SemanticKITTI -> nuScenes | mIoU: 46.7 -> 44.4 | Hurt |
| increase propagation distance from 0.30m to 0.60m (ablation) | 3DLabelProp · nuScenes -> SemanticKITTI | mIoU: 60.5 -> 57.2 | Hurt |
| reduce propagation distance from 0.30m to 0.10m (ablation) | 3DLabelProp · nuScenes -> SemanticKITTI | mIoU / FPS: mIoU 60.5 -> 60.2, FPS 1.2 -> 0.6 | No clear effect |
| choose 3DLabelProp hyperparameters differently (ablation) | 3DLabelProp · nuScenes -> SemanticKITTI | mIoU: only a difference of 4% between best and worst | No clear effect |
Closest papers by meaning
- Single Domain Generalization for LiDAR Semantic Segmentation (CVPR 2023 · similarity 0.94)
- Toward Enhanced Representation Learning for Single-Source Domain Generalization in LiDAR Semantic Segmentation (TPAMI 2026 · similarity 0.92)
- Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data (CVPR 2022 · similarity 0.92)
- Temporal Consistent 3D LiDAR Representation Learning for Semantic Perception in Autonomous Driving (CVPR 2023 · similarity 0.91)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.89)
- Geometry-Guided Domain Generalization for Monocular 3D Object Detection (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.
