Research map / Detection and segmentation / 3D detection and driving
ST3D++: Denoised Self-Training for Unsupervised Domain Adaptation on 3D Object Detection
Jihan Yang, Shaoshuai Shi, Zhe Wang, Hongsheng Li, Xiaojuan Qi
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 random object scaling to source pre-training (ablation) | source-only detector · Waymo->KITTI (car) | AP_3D: 27.48 -> 54.67 | Improved |
| add random object scaling before self-training (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 33.73 -> 65.64 | Improved |
| add triplet box partition to naive self-training (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 65.88 -> 67.39 | Improved |
| add memory ensemble and voting (MEV-C) (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 67.39 -> 70.17 | Improved |
| use hybrid criterion instead of IoU-only criterion (QAC) (ablation) | ST3D++ · Waymo->KITTI (pedestrian) | AP_3D: 51.92 -> 56.13 | Improved |
| use hybrid criterion instead of IoU-only criterion (QAC) (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 72.52 -> 72.52 | No clear effect |
| add curriculum data augmentation (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 72.52 -> 73.37 | Improved |
| add source-assisted self-denoised training (SASD) (ablation) | ST3D++ · Waymo->KITTI (car) | AP_3D: 73.37 -> 74.61 | Improved |
| add source-assisted self-denoised training (SASD) (ablation) | ST3D++ · Waymo->KITTI (cyclist) | AP_3D: 53.49 -> 60.76 | Improved |
| add source-assisted self-denoised training (SASD) (ablation) | ST3D++ · Waymo->KITTI (pedestrian) | AP_3D: 56.27 -> 59.21 | Improved |
| add target object size prior (SN) to ST3D++ | ST3D++ · Waymo->KITTI (car) | AP_3D: 65.64 -> 74.61 | Improved |
| add target object size prior (SN) to ST3D++ | ST3D++ · Waymo->Lyft (car) | AP_3D: 59.93 -> 58.86 | Hurt |
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 7 of 7 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
- ST3D: Self-Training for Unsupervised Domain Adaptation on 3D Object Detection (CVPR 2021 · similarity 0.96)
- Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection (CVPR 2024 · similarity 0.92)
- Revisiting Domain-Adaptive 3D Object Detection by Reliable, Diverse and Class-balanced Pseudo-Labeling (ICCV 2023 · similarity 0.91)
- Unsupervised Domain Adaptation for Monocular 3D Object Detection via Self-Training (ECCV 2022 · similarity 0.91)
- SP3D: Boosting Sparsely-Supervised 3D Object Detection via Accurate Cross-Modal Semantic Prompts (CVPR 2025 · similarity 0.88)
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · 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.
