Research map / Detection and segmentation / Open-set and few-shot detection
Better Regression Makes Better Test-time Adaptive 3D Object Detection
Jiakang Yuan, Bo Zhang, Kaixiong Gong, Xiangyu Yue, Botian Shi, Yu Qiao, Tao Chen
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 |
|---|---|---|---|
| adapt only the regression layers instead of BN layers (ablation) | SECOND-IoU test-time adaptation · Waymo -> KITTI | AP BEV / AP3D: 52.84 / 31.77 -> 59.38 / 45.54 | Improved |
| adapt regression layers instead of all parameters on nuScenes -> KITTI (ablation) | SECOND-IoU · nuScenes -> KITTI | AP BEV / AP3D: 33.23 / 19.60 -> 33.63 / 21.80 | Improved |
| apply existing TTA methods (TENT, SHOT, MATE) to 3D detection | SECOND-IoU · Waymo -> KITTI | AP BEV / AP3D: 46.64 / 28.83 -> 31.34 / 19.47 (TENT) | Hurt |
| add noise-consistency pseudo-label generation (NPG) (ablation) | naive pseudo-labels · Waymo -> KITTI | mean AP3D: 42.93 -> 44.91 | Improved |
| add confidence-guided regression refinement (CRR) (ablation) | naive pseudo-labels · Waymo -> KITTI | mean AP3D: 42.93 -> 45.61 | Improved |
| add class-balanced EMA updating (CBU) (ablation) | naive pseudo-labels + NPG + CRR · Waymo -> KITTI | mean AP3D: 47.97 -> 49.96 | Improved |
| use naive pseudo-labels without NPG/CRR/CBU (ablation) | source model · Waymo -> KITTI | mean AP3D: 42.01 -> 42.93 | Improved |
| vary noise inconsistency threshold tau and size threshold p (ablation) | Reg-TTA3D · Waymo -> KITTI | AP BEV / AP3D: tau 1.0/1.5/2.0: 49.12 / 49.96 / 48.88 AP3D | No clear effect |
| apply Reg-TTA3D in cross-weather settings | CenterPoint · nuScenes -> nuScenes-C Snow | mAP: 45.86 -> 47.09 | Improved |
| Reg-TTA3D vs UDA on cross-beam settings | Reg-TTA3D · nuScenes -> KITTI, Waymo -> nuScenes | AP3D: nuScenes->KITTI 23.13 vs ReDB 26.11 | Mixed |
Closest papers by meaning
- Revisiting Domain-Adaptive 3D Object Detection by Reliable, Diverse and Class-balanced Pseudo-Labeling (ICCV 2023 · similarity 0.91)
- Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection (CVPR 2024 · similarity 0.90)
- Test-Time Adaptive Object Detection with Foundation Model (NeurIPS 2025 · similarity 0.87)
- Unified Domain Generalization and Adaptation for Multi-View 3D Object Detection (NeurIPS 2024 · similarity 0.87)
- What How and When Should Object Detectors Update in Continually Changing Test Domains? (CVPR 2024 · similarity 0.87)
- Self-Bootstrapping for Versatile Test-Time Adaptation (ICML 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.
