Research map / Label-efficient and robust learning / Test-time adaptation
Uncertainty-Calibrated Test-Time Model Adaptation Without Forgetting
Mingkui Tan, Guohao Chen, Jiaxiang Wu, Yifan Zhang, Yaofo Chen, Peilin Zhao, Shuaicheng Niu
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 entropy-based reliable sample weighting S_ent (ablation) | EATA · ImageNet-C Gaussian noise (ResNet-50) level 5 | accuracy: 33.4 -> 49.6 | Improved |
| add diversity-based redundant sample filtering (ablation) | EATA · ImageNet-C Gaussian noise (ResNet-50) level 5 | accuracy / #backwards: 37,636 -> 28,168 backwards, 49.6 -> 50.4 | Improved |
| add active sample selection and Fisher regularizer (EATA vs Tent) | Tent · ImageNet-C Gaussian (ViT-Base) | accuracy: 33.4 -> 50.5 | Improved |
| add consistency loss and min-max entropy regularization (EATA-C vs EATA) | EATA · ImageNet-C avg (ViT-Base) | accuracy: 62.57 -> 66.65 | Improved |
| add consistency loss and min-max entropy regularization (EATA-C vs EATA) | EATA · ImageNet-C avg (ViT-Base) | ECE: 14.63 -> 5.14 | Improved |
| add consistency loss alone (ablation) | source model · ImageNet-C severity 5 single-domain avg (ResNet-50) | accuracy: 39.8 -> 47.7 | Improved |
| add min-max entropy regularizer (ablation) | EATA-C without ER · ImageNet-C severity 5 single-domain avg (ResNet-50) | accuracy / ECE: 48.8 -> 49.0, ECE 5.1 -> 4.6 | Improved |
| add Fisher regularizer (ablation) | ETA-C · ImageNet-C severity 5 single-domain avg (ResNet-50) | accuracy / ECE: 48.9 -> 49.0, ECE 5.4 -> 4.6 | Improved |
| add active sample selection (ablation) | EATA-C · ImageNet-C severity 5 single-domain avg (ResNet-50) | accuracy: 48.5 -> 49.0 | Improved |
| add min-max entropy regularizer to consistency loss (ablation) | EATA-C CL only · ImageNet-C severity 5 single-domain avg (ResNet-50) | accuracy: 47.7 -> 48.6 | Improved |
| use EATA instead of Tent (sample filtering) | Tent · ImageNet-C Gaussian (ResNet-50) | ECE: 11.7 -> 10.5 | Improved |
| add EATA-C over EATA on semantic segmentation | EATA · Cityscapes-to-ACDC lifelong | mIoU: 57.0 -> 61.6 | Improved |
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 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
- Unified Entropy Optimization for Open-Set Test-Time Adaptation (CVPR 2024 · similarity 0.90)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.90)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.90)
- AETTA: Label-Free Accuracy Estimation for Test-Time Adaptation (CVPR 2024 · similarity 0.90)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.89)
- Bayesian Weight Enhancement with Steady-State Adaptation for Test-time Adaptation in Dynamic Environments (ICML 2025 · similarity 0.89)
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.
