Research map / Label-efficient and robust learning / Test-time adaptation
Improving Batch Normalization with Test-Time Adaptation for Robust Object Detection in Self-Driving
Dacheng Liao, Mengshi Qi, Liang Liu, Huadong Ma
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 GCM loss to LearnableBN with EM loss (ablation) | LearnableBN (BEVFormer) · nuScenes-C Snow | NDS: 0.2509 -> 0.2687 | Improved |
| add GCM loss to LearnableBN with EM loss (ablation) | LearnableBN (BEVFormer) · nuScenes-C Brightness | NDS: 0.4583 -> 0.4680 | Improved |
| add semantic-consistency dual-stage adaptation (ablation) | LearnableBN (BEVFormer) · nuScenes-C Brightness | NDS: 0.4680 -> 0.4835 | Improved |
| add semantic-consistency dual-stage adaptation (ablation) | LearnableBN (BEVFormer) · nuScenes-C Color Quant | NDS: 0.4099 -> 0.4192 | Improved |
| add semantic-consistency dual-stage adaptation (ablation) | LearnableBN (BEVFormer) · nuScenes-C Snow | NDS: 0.2687 -> 0.2718 | Improved |
| use entropy minimization loss only in LearnableBN (ablation) | LearnableBN (BEVFormer) · nuScenes-C (six corruptions) | NDS: Snow 0.2297 -> 0.2509; Brightness 0.4908 -> 0.4583 | Mixed |
| replace EMA BN statistics with learnable mixture coefficient (LearnableBN) vs ReviseBN | BEVFormer R101 · nuScenes-C Low Light | NDS: 0.2631 -> 0.3671 | Improved |
| apply TENT (entropy minimization on BN affine params) to BEVFormer | BEVFormer R101 · nuScenes-C Low Light | NDS: 0.4011 -> 0.2636 (easy) | Hurt |
| apply AdaBn and ARM BN adaptation at batch size 1 | BEVFormer R101 · nuScenes-C Low Light | NDS: 0.3212 -> 0.0772 (AdaBn) | Hurt |
| apply LearnableBN to Sparse4D | Sparse4D R101 · nuScenes-C Motion Blur | NDS: 0.3563 -> 0.4035 | Improved |
| apply LearnableBN to MonoFlex on KITTI-C | MonoFlex · KITTI-C Pedestrian | AP3D|R40: Contrast 2.08 -> 14.74; Fog 2.25 -> 9.62 | Mixed |
| apply LearnableBN when baseline already strong (fog) | BEVFormer R101 · nuScenes-C Fog | NDS: 0.4796 -> 0.4816 | No clear effect |
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.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 4 of 4 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
- SoTTA: Robust Test-Time Adaptation on Noisy Data Streams (NeurIPS 2023 · similarity 0.88)
- NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (NeurIPS 2022 · similarity 0.87)
- CodeMerge: Codebook-Guided Model Merging for Robust Test-Time Adaptation in Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- Test-Time Adaptive Object Detection with Foundation Model (NeurIPS 2025 · similarity 0.86)
- Uncertainty-Guided Never-Ending Learning to Drive (CVPR 2024 · similarity 0.85)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.85)
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.
