Research map / Image generation / Image restoration
Modulating Image Restoration With Continual Levels via Adaptive Feature Modification Layers
Jingwen He, Chao Dong, Yu Qiao
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 AdaFM feature-modification layers and adapt only those layers to the target level | basic CNN restoration model · Set5 ×3→×4 | PSNR (dB): 32.13 -> 32.00 | Improved |
| Use AdaFM-Net instead of AdaBN | restoration modulation · CBSD68 σ75 | PSNR (dB): 19.68 -> 26.35 | Improved |
| Use AdaFM-Net instead of Conditional IN | restoration modulation · Set5 ×4 | PSNR (dB): 31.30 -> 32.00 | Improved |
| Increase AdaFM filter size from 1×1 to 3×3 (ablation) | AdaFM-Net · Set5 ×3→×4 | PSNR (dB): 31.42 -> 31.88 | Improved |
| Increase AdaFM filter size from 5×5 to 7×7 (ablation) | AdaFM-Net · Set5 ×3→×4 | PSNR (dB): 32.00 -> 32.03 | Improved |
| Add AdaFM layers after activations rather than before activations (ablation) | AdaFM-Net · Set5 ×3→×4 | PSNR (dB): 32.00 -> 31.84 | Hurt |
| Insert AdaFM layers in residual blocks only rather than after all convolution layers (ablation) | AdaFM-Net · Set5 ×3→×4 | PSNR (dB): 32.01 -> 32.00 | No clear effect |
| Adapt from an easier restoration level to a harder level (ablation) | AdaFM-Net · DeJPEG q80→q10 versus q10→q80 | PSNR distance (dB): 1.04 -> 0.20 | Improved |
| Reduce adaptation range from q80→q10 to q80→q60 (ablation) | AdaFM-Net · LIVE1 DeJPEG | PSNR distance (dB): 0.20 -> 0.02 | Improved |
| Reduce adaptation range from σ15→σ75 to σ15→σ25 (ablation) | AdaFM-Net · CBSD68 denoising | PSNR distance (dB): 0.14 -> 0.01 | Improved |
| Reduce super-resolution adaptation range from ×2→×4 to ×2→×3 (ablation) | AdaFM-Net · Set5 super-resolution | PSNR distance (dB): 0.47 -> 0.36 | Improved |
| Increase AdaFM filter size from 1×1 to 7×7 (ablation) | AdaFM-Net · CBSD68 σ15→σ75 | PSNR (dB): 26.35 -> 26.40 | No clear effect |
Closest papers by meaning
- On-Demand Learning for Deep Image Restoration (ICCV 2017 · similarity 0.87)
- Denoising Prior Driven Deep Neural Network for Image Restoration (TPAMI 2018 · similarity 0.87)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
- Zero-Shot Image Restoration Using Few-Step Guidance of Consistency Models (and Beyond) (CVPR 2025 · similarity 0.86)
- Boosting Image Restoration via Priors from Pre-trained Models (CVPR 2024 · similarity 0.86)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 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.
