Research map / Point clouds / Segmentation and detection
Rethinking LiDAR Domain Generalization: Single Source as Multiple Density Domains
Jaeyeul Kim, Jungwan Woo, Jeonghoon Kim, Sunghoon Im
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 point-voxel feature encoding (no intensity, voxel-wise encoding) (ablation) | MinkowskiNet baseline · SemanticKITTI->nuScenes | mIoU: 40.7 -> 43.0 | Improved |
| add density-aware embedding module (ablation) | DDFE with point-voxel encoding · nuScenes->SemanticKITTI | mIoU: 35.0 -> 40.5 | Improved |
| add density soft clipping (ablation) | DDFE (a)+(b) · SemanticKITTI->nuScenes | mIoU: 45.7 -> 46.2 | Improved |
| add density soft clipping (ablation) | DDFE (a)+(b) · nuScenes->SemanticKITTI | mIoU: 40.5 -> 41.8 | Improved |
| add density augmentation (enhanced-Mix3D + beam sampling) (ablation) | DDFE · SemanticKITTI->nuScenes | mIoU: 46.2 -> 50.1 | Improved |
| add density augmentation (ablation) | DDFE · nuScenes->SemanticKITTI | mIoU: 41.8 -> 46.3 | Improved |
| replace Mix3D with enhanced-Mix3D (translation and rotation) (ablation) | DDFE · SemanticKITTI->nuScenes | mIoU: 47.4 -> 48.5 | Improved |
| add beam sampling to Mix3D (ablation) | DDFE + Mix3D · nuScenes->SemanticKITTI | mIoU: 44.8 -> 44.2 | Hurt |
| add beam sampling to enhanced-Mix3D (ablation) | DDFE + E-Mix3D · nuScenes->SemanticKITTI | mIoU: 46.4 -> 46.5 | No clear effect |
| use voxel size 20cm instead of 5cm (ablation) | DDFE · SemanticKITTI->nuScenes (MinkNet42) | mIoU: 48.6 -> 50.1 (K->N); 51.3 -> 46.3 (N->K) | Mixed |
| apply DDFE over DGLSS on unseen domains | MinkowskiNet · Waymo->SemanticKITTI | mIoU: 51.23 -> 57.07 | Improved |
| add DDFE module to backbone | MinkNet42 · nuScenes inference time | inference time (ms): 36ms -> 44ms | Hurt |
Closest papers by meaning
- GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (CVPR 2025 · similarity 0.88)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.87)
- D^3CTTA: Domain-Dependent Decorrelation for Continual Test-Time Adaption of 3D LiDAR Segmentation (CVPR 2025 · similarity 0.86)
- LiT: Unifying LiDAR "Languages" with LiDAR Translator (NeurIPS 2024 · similarity 0.86)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.86)
- Unlocking Generalization Power in LiDAR Point Cloud Registration (CVPR 2025 · similarity 0.86)
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.
