Research map / Detection and segmentation / 3D detection and driving
DetMatch: Two Teachers Are Better than One for Joint 2D and 3D Semi-Supervised Object Detection
Jinhyung Park, Chenfeng Xu, Yiyang Zhou, Masayoshi Tomizuka, Wei Zhan
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 |
|---|---|---|---|
| replace confidence thresholding with 2D-3D teacher matching (ablation) | DetMatch · KITTI 1% 3D | mAP: 54.4 -> 57.9 | Improved |
| add 2D-teacher/3D-student box consistency (ablation) | DetMatch · KITTI 1% 3D | mAP: 57.9 -> 59.4 | Improved |
| add 2D-teacher/3D-student class consistency (ablation) | DetMatch · KITTI 1% 2D | mAP: 69.8 -> 71.4 | Improved |
| use MSE instead of Focal loss for class consistency (ablation) | DetMatch · KITTI 1% 3D | mAP: 59.0 -> 58.2 | Hurt |
| set tau_hung to -1.5 instead of -1 or -2 (ablation) | DetMatch · KITTI 1% 3D | mAP: 54.2/52.4 -> 57.9 | Improved |
| remove COCO pre-training of 2D detector (ablation) | DetMatch · KITTI 1% 2D | mAP: 71.4 -> 59.1 | Hurt |
| remove COCO pre-training of 2D detector (ablation) | DetMatch · KITTI 1% 3D | Pedestrian AP: 57.3 -> 55.3 | No clear effect |
| apply 2D confidence thresholding (Unbiased Teacher) SSL | Confidence Thresholding · KITTI 1% 2D | mAP: 65.3 -> 60.4 | Hurt |
| DetMatch over 3D confidence thresholding | DetMatch · KITTI 1% 3D | Pedestrian AP: 42.7 -> 57.3 | Improved |
| DetMatch over 3D confidence thresholding | DetMatch · Waymo 1% 3D | Car L1 mAP: 52.6 -> 52.2 | Hurt |
| DetMatch over 3D confidence thresholding | DetMatch · Waymo 1% 3D | Ped L1 mAP: 35.2 -> 39.5 | Improved |
| swap PV-RCNN for SECOND | DetMatch · KITTI 1% 3D | mAP: 38.8 -> 49.4 | Improved |
Later papers that compared with this method
1 later accepted paper reports DetMatch as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Hierarchical Supervision and Shuffle Data Augmentation for 3D Semi-Supervised Object Detection (CVPR 2023 · better on 5 of 5 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 3DIoUMatch — 3DIoUMatch: Leveraging IoU Prediction for Semi-Supervised 3D Object Detection (CVPR 2021) · better on 3 of 3 shared results
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
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.91)
- Dual-Perspective Knowledge Enrichment for Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.90)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.89)
- Semi-supervised 3D Object Detection with PatchTeacher and PillarMix (AAAI 2024 · similarity 0.88)
- Decoupled Pseudo-labeling for Semi-Supervised Monocular 3D Object Detection (CVPR 2024 · similarity 0.87)
- Sparse Semi-DETR: Sparse Learnable Queries for Semi-Supervised Object Detection (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.
