Research map / Image generation / Image restoration
Rethinking Coarse-To-Fine Approach in Single Image Deblurring
Sung-Jin Cho, Seo-Won Ji, Jun-Pyo Hong, Seung-Won Jung, Sung-Jea Ko
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 multi-output single decoder (MOSD) (ablation) | Baseline U-Net · GoPro test, component ablation | PSNR (dB): 31.16 -> 31.33 | Improved |
| add multi-input single encoder (MISE) alone (ablation) | Baseline U-Net · GoPro test, component ablation | PSNR (dB): 31.16 -> 31.17 | No clear effect |
| add MISE on top of MOSD (ablation) | MOSD · GoPro test, component ablation | PSNR (dB): 31.33 -> 31.38 | Improved |
| add asymmetric feature fusion (AFF) (ablation) | Baseline U-Net · GoPro test, component ablation | PSNR (dB): 31.16 -> 31.33 | Improved |
| add MISE on top of AFF (ablation) | AFF · GoPro test, component ablation | PSNR (dB): 31.33 -> 31.39 | Improved |
| add MISE, MOSD and AFF together (ablation) | Baseline U-Net · GoPro test, component ablation | PSNR (dB): 31.16 -> 31.46 | Improved |
| add multi-scale frequency reconstruction (MSFR) loss (ablation) | MISE+MOSD+AFF · GoPro test, component ablation | PSNR (dB): 31.46 -> 31.73 | Improved |
| use FAM instead of concatenation for fusing multi-scale input features (ablation) | MIMO-UNet · GoPro test, MIMO-UNet fusion ablation | PSNR (dB): 31.66 -> 31.73 | Improved |
| use FAM instead of element-wise sum (ablation) | MIMO-UNet · GoPro test, MIMO-UNet fusion ablation | PSNR (dB): 31.60 -> 31.73 | Improved |
| increase residual blocks per encoder/decoder block from 8 to 20 (MIMO-UNet+) | MIMO-UNet · GoPro test | PSNR (dB): 31.73 -> 32.45 | Improved |
| add geometric self-ensemble (MIMO-UNet++) | MIMO-UNet+ · GoPro test | PSNR (dB): 32.45 -> 32.68 | Improved |
| add geometric self-ensemble (MIMO-UNet++) | MIMO-UNet+ · RealBlur test | PSNR (dB): 31.92 -> 32.05 | Improved |
Closest papers by meaning
- XYDeblur: Divide and Conquer for Single Image Deblurring (CVPR 2022 · similarity 0.88)
- Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring (CVPR 2024 · similarity 0.86)
- Leveraging Phase Information to Boost Unrolled Network Learning for Image Deblurring (ECCV 2026 · similarity 0.86)
- Efficient RAW Image Deblurring with Adaptive Frequency Modulation (NeurIPS 2025 · similarity 0.84)
- Blur2Blur: Blur Conversion for Unsupervised Image Deblurring on Unknown Domains (CVPR 2024 · similarity 0.84)
- Rethinking Multi-Scale Representations in Deep Deraining Transformer (AAAI 2024 · 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.
