Research map / Detection and segmentation / 3D detection and driving
LEROjD: Lidar Extended Radar-Only Object Detection
Patrick Palmer, Martin Krüger, Stefan Schütte, Richard Altendorfer, Ganesh Adam, Torsten Bertram
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 multi-stage training with mixed radar+lidar and voxel thin-out before radar-only fine-tuning | PointPillars radar-only · VoD (SR mAP) | mAP: 36.7 -> 39.7 | Improved |
| add multi-stage training with mixed radar+lidar and voxel thin-out before radar-only fine-tuning | PointPillars radar-only · VoD (MR mAP) | mAP: 11.9 -> 15.4 | Improved |
| use lidar-only multi-stage pre-training with random thin-out (ablation) | PointPillars radar-only · VoD | mAP SR / MR: SR 36.7 -> 36.4, MR 11.9 -> 16.1 | Mixed |
| mix radar points into lidar pre-training stages instead of lidar only (voxel thin-out) (ablation) | MSTM · VoD (SR mAP) | mAP: 34.1 -> 39.7 | Improved |
| mix radar points into lidar pre-training stages instead of lidar only (random thin-out) (ablation) | MSTM · VoD (SR mAP) | mAP: 36.4 -> 35.6 | Hurt |
| use multi-stage instead of single-stage training on thinned-out lidar only (ablation) | lidar-only detector · VoD lidar thin-out | mAP | No clear effect |
| use multi-stage instead of single-stage training on mixed radar+lidar with knn or voxel thin-out (ablation) | mixed radar+lidar detector · VoD thin-out stages | mAP | Improved |
| initialize radar student with teacher RL SSTM 1 weights | PointPillars radar-only · VoD (SR mAP) | mAP: 36.7 -> 39.0 | Improved |
| initialize radar student with lidar-only teacher weights | PointPillars radar-only · VoD (SR mAP) | mAP: 36.7 -> 34.9 | Hurt |
| add logit-KD with lidar-only teacher (ablation) | PointPillars radar-only · VoD (SR mAP) | mAP: 34.9 -> 32.8 | Hurt |
| add feature-KD with mixed teacher RL SSTM 1/4 vox (ablation) | student initialized from teacher · VoD (MR mAP) | mAP: 15.3 -> 15.8 | Improved |
| use joint KD with all three losses (ablation) | student, teacher RL SSTM 1/4 vox · VoD (MR mAP) | mAP: 15.3 -> 9.7 | Hurt |
Closest papers by meaning
- RadarDistill: Boosting Radar-based Object Detection Performance via Knowledge Distillation from LiDAR Features (CVPR 2024 · similarity 0.90)
- SCKD: Semi-Supervised Cross-Modality Knowledge Distillation for 4D Radar Object Detection (AAAI 2025 · similarity 0.90)
- CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation (CVPR 2024 · similarity 0.89)
- Towards Robust 3D Object Detection with LiDAR and 4D Radar Fusion in Various Weather Conditions (CVPR 2024 · similarity 0.88)
- L4DR: LiDAR-4DRadar Fusion for Weather-Robust 3D Object Detection (AAAI 2025 · similarity 0.88)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.87)
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.
