Research map / Label-efficient and robust learning / Test-time adaptation
Efficient Diffusion-Driven Corruption Editor for Test-Time Adaptation
Yeongtak Oh, Jonghyun Lee, Jooyoung Choi, Dahuin Jung, Uiwon Hwang, Sungroh Yoon
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 |
|---|---|---|---|
| combine PIXMIX and SimSiam augmentations in corruption modeling (ablation) | Decorruptor-DPM · ImageNet-C ResNet-50 | accuracy (%): 28.2 -> 30.5 | Improved |
| ensemble 4 edited images from consistency model instead of single edit (ablation) | Decorruptor-CM · ImageNet-C ResNet-50 | accuracy (%): 30.5 -> 32.8 | Improved |
| increase CM ensemble from 4 to 8 edits (ablation) | Decorruptor-CM · ImageNet-C ResNet-50 | accuracy (%): 32.8 -> 34.2 | Improved |
| use single-edit CM instead of DPM (ablation) | Decorruptor · ImageNet-C | accuracy (%): about 2-3% lower | Hurt |
| replace pixel-space DDA with latent-diffusion Decorruptor | diffusion TTA · ImageNet-C ResNet-50 | runtime (s/sample): 19.5 -> 0.42 | Improved |
| apply DDA to Swin-T / ConvNeXt-T on ImageNet-C-bar | DDA · ImageNet-C-bar Swin-T | accuracy (%): 44.2 -> 43.8 | Hurt |
| apply DDA on top of PIXMIX source model on ImageNet-A | DDA · ImageNet-A | accuracy (%): 6.3 -> 5.2 | Hurt |
| add Decorruptor-DPM to PIXMIX source model on ImageNet-A | PIXMIX · ImageNet-A | accuracy (%): 6.3 -> 8.1 | Improved |
| apply MEMO on stronger architectures | MEMO · ImageNet-C Swin-T | accuracy (%): 33.1 -> 29.5 | Hurt |
Closest papers by meaning
- Test-Time Degradation Adaptation for Open-Set Image Restoration (ICML 2024 · similarity 0.90)
- High-Fidelity Diffusion-Based Image Editing (AAAI 2024 · similarity 0.87)
- TiNO-Edit: Timestep and Noise Optimization for Robust Diffusion-Based Image Editing (CVPR 2024 · similarity 0.87)
- Everything to the Synthetic: Diffusion-driven Test-time Adaptation via Synthetic-Domain Alignment (CVPR 2025 · similarity 0.87)
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.86)
- PTTA: Purifying Malicious Samples for Test-Time Model Adaptation (ICML 2025 · similarity 0.86)
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.
