Research map / Label-efficient and robust learning / Test-time adaptation
ActMAD: Activation Matching To Align Distributions for Test-Time-Training
Muhammad Jehanzeb Mirza, Pol Jané Soneira, Wei Lin, Mateusz Kozinski, Horst Possegger, Horst Bischof
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 |
|---|---|---|---|
| Align individual spatial activations across multiple layers rather than only the final layer (ablation) | ActMAD · KITTI-Fog | mAP@50: 36.0 -> 47.7 | Improved |
| Align individual pixel statistics rather than channel-averaged statistics (ablation) | ActMAD · KITTI-Fog | mAP@50: 38.6 -> 47.7 | Improved |
| Use mean and variance alignment instead of central moment difference (ablation) | ActMAD · KITTI-Fog | mAP@50: 41.2 -> 47.7 | Improved |
| Use mean and variance alignment instead of central moment difference (ablation) | ActMAD · CIFAR-10C | Top-1 error (%): 10.5 -> 10.4 | No clear effect |
| Update all parameters rather than only affine parameters (ablation) | ActMAD · KITTI-Fog | mAP@50: 45.2 -> 47.7 | Improved |
| Update all parameters rather than only affine parameters (ablation) | ActMAD · CIFAR-10C | Top-1 error (%): 10.6 -> 10.4 | No clear effect |
| Reduce the adaptation batch size from 125 to 10 (ablation) | ActMAD · ImageNet-C level 5 | Top-1 error (%): 66.0 -> 72.8 | Hurt |
| Adapt with location-aware activation matching | WRN-40-2 · CIFAR-10C level 5 | Top-1 error (%): 18.3 -> 10.4 | Improved |
| Adapt with location-aware activation matching | ViT-B/16 · CIFAR-100C level 5 | Top-1 error (%): 35.1 -> 22.4 | Improved |
| Adapt with location-aware activation matching | ResNet-18 · ImageNet-C level 5 | Top-1 error (%): 86.2 -> 66.0 | Improved |
| Use ActMAD rather than CoTTA for continuous corruption adaptation | WRN-40-2 · CIFAR-10C level 5, continuous adaptation | Top-1 error (%): 11.5 -> 11.4 | No clear effect |
| Use ActMAD rather than CoTTA for continuous corruption adaptation | WRN-40-2 · CIFAR-100C level 5, continuous adaptation | Top-1 error (%): 35.7 -> 35.9 | No clear effect |
Later papers that compared with this method
4 later accepted papers report ActMAD as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- CD-Buffer: Complementary Dual-Buffer Framework for Test-Time Adaptation in Adverse Weather Object Detection (CVPR 2026 · better on 5 of 5 shared results)
- Multi-Task Test-time Adaptation via Gradient Consensus and Plasticity Constraint (AAAI 2026 · better on 2 of 2 shared results)
- Synchronizing Task Behavior: Aligning Multiple Tasks during Test-Time Training (ICCV 2025 · better on 3 of 4 shared results)
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · better on 1 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
- AcTTA: Rethinking Test-Time Adaptation via Dynamic Activation (CVPR 2026 · similarity 0.86)
- Extremely Simple Activation Shaping for Out-of-Distribution Detection (ICLR 2023 · similarity 0.86)
- What How and When Should Object Detectors Update in Continually Changing Test Domains? (CVPR 2024 · similarity 0.84)
- Self-Bootstrapping for Versatile Test-Time Adaptation (ICML 2025 · similarity 0.84)
- Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement (ICLR 2024 · similarity 0.84)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.84)
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.
