Research map / Detection and segmentation / 3D detection and driving
Revisiting Domain-Adaptive 3D Object Detection by Reliable, Diverse and Class-balanced Pseudo-Labeling
Zhuoxiao Chen, Yadan Luo, Zheng Wang, Mahsa Baktashmotlagh, Zi Huang
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 cross-domain examination (CDE) to pseudo-label filtering (ablation) | ReDB · Waymo -> KITTI (Moderate) | mAP 3D: 47.66 -> 50.10 | Improved |
| add OBC-based diversity downsampling (ablation) | ReDB · Waymo -> KITTI (Moderate) | mAP 3D: 47.07 -> 50.10 | Improved |
| add source ground-truth object sampling to target point clouds (ablation) | ReDB baseline without sampling · Waymo -> KITTI (Moderate) | mAP 3D: 42.26 -> 45.21 | Improved |
| add target pseudo-label sampling to target point clouds (ablation) | ReDB baseline without sampling · Waymo -> KITTI (Moderate) | mAP 3D: 42.26 -> 46.04 | Improved |
| source GT sampling alone improves easy level much more than moderate and hard (ablation) | ReDB baseline · Waymo -> KITTI | mAP 3D: Easy 48.41 -> 54.10; Moderate 42.26 -> 45.21; Hard 40.18 -> 42.42 | Mixed |
| replace ST3D++ with ReDB on PointRCNN backbone | PointRCNN · nuScenes -> KITTI (Moderate) | average AP 3D: 36.58 -> 45.05 | Improved |
| replace ST3D++ with ReDB on large environmental gap | SECOND · nuScenes -> KITTI (Moderate) | mean AP 3D: 21.64 -> 26.11 | Improved |
| increase downsampling rate d to 10 (ablation) | ReDB · Waymo -> KITTI | mAP 3D: -1.37% mAP3D, -2.57% mAP BEV | Hurt |
| vary CDE IoU threshold between 0.3 and 0.6 (ablation) | ReDB · Waymo -> KITTI | mAP 3D and BEV: 2.69% and 2.07% fluctuation | No clear effect |
| ReDB versus SN on Waymo -> KITTI car and pedestrian 3D AP | ReDB · Waymo -> KITTI (Moderate) | Car AP 3D: SN 54.14 vs ReDB 54.12 | Mixed |
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.
- ST3D++ — ST3D: Self-Training for Unsupervised Domain Adaptation on 3D Object Detection (CVPR 2021) · better on 11 of 11 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
- Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection (CVPR 2024 · similarity 0.93)
- Unsupervised Domain Adaptive 3D Detection With Multi-Level Consistency (ICCV 2021 · similarity 0.92)
- ST3D: Self-Training for Unsupervised Domain Adaptation on 3D Object Detection (CVPR 2021 · similarity 0.91)
- Better Regression Makes Better Test-time Adaptive 3D Object Detection (ECCV 2024 · similarity 0.91)
- Unified Domain Generalization and Adaptation for Multi-View 3D Object Detection (NeurIPS 2024 · similarity 0.89)
- CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 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.
