Research map / Point clouds / Segmentation and detection
3D Object Detection with a Self-Supervised Lidar Scene Flow Backbone
Emeç Erçelik, Ekim Yurtsever, Mingyu Liu, Zhijie Yang, Hanzhen Zhang, Pınar Topçam, Maximilian Listl, Yılmaz Kaan Çaylı, Alois Knoll
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 |
|---|---|---|---|
| initialize backbone with self-supervised scene flow pre-training | Point-GNN · KITTI val car | 3D AP Hard: 77.70 -> 80.12 | Improved |
| initialize backbone with self-supervised scene flow pre-training | PointPillars · KITTI val car | 3D AP Hard: 67.76 -> 74.32 | Improved |
| initialize backbone with self-supervised scene flow pre-training | PointPillars · KITTI test car | 3D AP Mod: 68.57 -> 72.99 | Improved |
| initialize backbone with self-supervised scene flow pre-training | PointPillars · KITTI test car | BEV AP Easy: 90.74 -> 88.92 | Hurt |
| initialize backbone with self-supervised scene flow pre-training | PointPillars · nuScenes val | mAP: 40.02 -> 42.06 | Improved |
| initialize backbone with self-supervised scene flow pre-training | CenterPoint · nuScenes val | mAP: 49.13 -> 49.94 | Improved |
| initialize backbone with self-supervised scene flow pre-training | CenterPoint · nuScenes test | mAP: 49.54 -> 51.42 | Improved |
| pre-train with scene flow when only 1% labeled data used (ablation) | Point-GNN · KITTI val car (1% labels) | 3D AP Easy: 63.34 -> 66.47 | Improved |
| pre-train with scene flow when 20% labeled data used (ablation) | Point-GNN · KITTI val car (20% labels) | 3D AP Easy: 88.47 -> 88.65 | No clear effect |
| add alternating training between flow and detection (ablation) | Point-GNN · KITTI val car | 3D AP Hard: 2.32% | Improved |
| use scene flow pre-training instead of GCC3D | CenterPoint · nuScenes val (5% labels) | mAP: 32.75 -> 36.04 | Improved |
| use scene flow pre-training instead of PointContrast | CenterPoint · nuScenes val (10% labels) | mAP: 38.25 -> 41.29 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · better on 10 of 10 shared results
- Point-GNN — Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LISO: Lidar-only Self-Supervised 3D Object Detection (ECCV 2024 · similarity 0.91)
- SLIM: Self-Supervised LiDAR Scene Flow and Motion Segmentation (ICCV 2021 · similarity 0.91)
- Towards A Weakly Supervised Framework for 3D Point Cloud Object Detection and Annotation (TPAMI 2021 · similarity 0.90)
- SeMoLi: What Moves Together Belongs Together (CVPR 2024 · similarity 0.90)
- Motion Cues from Image-based Point Tracking for LiDAR Scene Flow Estimation (NeurIPS 2026 · similarity 0.89)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · 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.
