Research map / Detection and segmentation / Weakly supervised detection and segmentation
CSOT: Cross-Scan Object Transfer for Semi-Supervised LiDAR Object Detection
Jinglin Zhan, Tiejun Liu, Rengang Li, Zhaoxiang Zhang, Yuntao Chen
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 CSOT (HotspotNet-guided object transfer with spatial-aware loss) to 5% labeled CenterPoint (ablation) | CenterPoint on nuScenes · nuScenes val (5% labels) | mAP: 43.8 -> 50.4 | Improved |
| add one cycle of self-training on top of CSOT (ablation) | CenterPoint on nuScenes · nuScenes val (5% labels) | mAP: 50.4 -> 56.4 | Improved |
| add CSOT to 1% labeled CenterPoint on Waymo (ablation) | CenterPoint on Waymo · Waymo val (1% labels) | mAPH: 63.7 -> 66.9 | Improved |
| replace GT-Aug with learned object transfer for fully supervised detector (ablation) | CenterPoint on Waymo (100% labels) · Waymo val | mAPH: 70.2 -> 73.1 | Improved |
| replace HotspotNet object transfer with GT-Aug, CA-Aug or LiDAR-Aug in CSOT (ablation) | CSOT on nuScenes · nuScenes val (5% labels, with self-training) | mAP: 56.4 -> 49.7 | Hurt |
| use learned placement scoring instead of rule-based (ablation) | CSOT with HotspotNet · nuScenes val (5% labels, with self-training) | mAP: 53.6 -> 56.4 | Improved |
| remove orientation constraint from fitness scores (ablation) | HotspotNet scorer · nuScenes val (5% labels) | mAP: 56.4 -> 55.1 | Hurt |
| add orientation constraint to category consistency score (ablation) | HotspotNet scorer · nuScenes val (5% labels) | mAP: 51.6 -> 52.2 | Improved |
| apply CSOT to pillar-based CenterPoint (PFE) (ablation) | CenterPoint PFE · nuScenes val (5% labels) | mAP: 30.2 -> 40.0 | Improved |
| apply CSOT and self-training to anchor-based SECOND (ablation) | SECOND · nuScenes val (5% labels) | mAP: 38.4 -> 48.6 | Improved |
| repeat self-training iteratively instead of one cycle (ablation) | CSOT · nuScenes | mAP | No clear effect |
Closest papers by meaning
- SSDA3D: Semi-supervised Domain Adaptation for 3D Object Detection from Point Cloud (AAAI 2023 · similarity 0.88)
- MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection (ICLR 2024 · similarity 0.88)
- A-Teacher: Asymmetric Network for 3D Semi-Supervised Object Detection (CVPR 2024 · similarity 0.87)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.87)
- SeMoLi: What Moves Together Belongs Together (CVPR 2024 · similarity 0.86)
- GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (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.
