Research map / Label-efficient and robust learning / Test-time adaptation
Hybrid-TTA: Continual Test-time Adaptation via Dynamic Domain Shift Detection
Hyewon Park, Hyejin Park, Jueun Ko, Dongbo Min
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 MIMA to full tuning (ablation) | Full Tuning · Cityscapes-to-ACDC (10 rounds) | mean mIoU: 59.3 -> 61.9 | Improved |
| add DDSD to full tuning (ablation) | Full Tuning · Cityscapes-to-ACDC (10 rounds) | mean mIoU: 59.3 -> 59.8 | Improved |
| add DDSD to full tuning (ablation) | Full Tuning · Cityscapes-to-ACDC (10 rounds) | FPS: 1.976 -> 2.546 | Improved |
| add DDSD and MIMA together (ablation) | Full Tuning · Cityscapes-to-ACDC (10 rounds) | mean mIoU: 59.3 -> 62.2 | Improved |
| add DDSD on top of FT+MIMA over 20 rounds (ablation) | FT+MIMA · Cityscapes-to-ACDC (20 rounds) | mean mIoU: 59.6 -> 61.5 | Improved |
| add DDSD to full tuning over 20 rounds (ablation) | Full Tuning · Cityscapes-to-ACDC (20 rounds) | mean mIoU: 55.9 -> 58.9 | Improved |
| add reconstruction loss weight 0.3 vs none (ablation) | FT+MIMA · Cityscapes-to-ACDC (10 rounds) | mIoU: 61.3 -> 61.9 | Improved |
| raise reconstruction loss weight to 1.0 (ablation) | FT+MIMA · Cityscapes-to-ACDC (10 rounds) | mIoU: 61.9 -> 60.6 | Hurt |
| use dynamic DDSD detector instead of fixed-threshold ADD (ablation) | Hybrid-TTA · OnDA rain benchmark | mean mIoU: 64.7 -> 66.73 | Improved |
| always use FT instead of DDSD switching (with MIMA) (ablation) | Hybrid-TTA · Cityscapes-to-ACDC (10 rounds) | mean mIoU: 62.2 -> 61.9 | Hurt |
| use only adapter tuning (with MIMA) (ablation) | Hybrid-TTA · Cityscapes-to-ACDC (10 rounds) | mean mIoU: 62.2 -> 59.27 | Hurt |
| add MIMA to adapter tuning (ablation) | Adapter Tuning · OnDA rain benchmark (target mean) | mIoU: 60.5 -> 65.3 | Improved |
Closest papers by meaning
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.91)
- PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation (AAAI 2025 · similarity 0.89)
- ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (ICLR 2024 · similarity 0.89)
- DPCore: Dynamic Prompt Coreset for Continual Test-Time Adaptation (ICML 2025 · similarity 0.88)
- A Versatile Framework for Continual Test-Time Domain Adaptation: Balancing Discriminability and Generalizability (CVPR 2024 · similarity 0.88)
- Continual-MAE: Adaptive Distribution Masked Autoencoders for Continual Test-Time Adaptation (CVPR 2024 · 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.
