Research map / Image generation / Image restoration
Intriguing Findings of Frequency Selection for Image Deblurring
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Res FFT-ReLU blocks | NAFNet64 · GoPro test (Group I) | PSNR: 33.69 -> 33.85 | Helped |
| Add Res FFT-ReLU blocks | MIMO-UNet+ · HIDE test (Group I) | PSNR: 29.99 -> 31.66 | Helped |
| Add FFT-ReLU stream alongside the spatial residual stream (ablation) | RSNet · GoPro test (Group I) | PSNR: 28.06 -> 30.30 | Helped |
| Use only the FFT-ReLU stream and remove the spatial residual stream (ablation) | RSNet · GoPro test (Group I) | PSNR: 28.06 -> 25.64 | Hurt |
| Replace complex 1×1 convolution weights with real weights (ablation) | RSNet with FFT-ReLU stream · GoPro test (Group I) | PSNR: 30.30 -> 29.91 | Hurt |
| Concatenate real and imaginary FFT parts and use real convolution weights (ablation) | RSNet with FFT-ReLU stream · GoPro test (Group I) | PSNR: 30.30 -> 30.32 | No clear effect |
| Replace 1×1 frequency convolutions with 3×3 convolutions (ablation) | RSNet with FFT-ReLU stream · GoPro test (Group I) | PSNR: 30.32 -> 31.15 | Helped |
| Replace ReLU with GeLU in the FFT-ReLU stream (ablation) | RSNet with FFT-ReLU stream · GoPro test (Group I) | PSNR: 30.30 -> 30.42 | Helped |
| Use inverse ReLU frequency selection instead of ReLU (ablation) | RSNet with simple FFT-ReLU stream · GoPro test (Group I) | PSNR: 29.51 -> 29.47 | No clear effect |
| Set a positive frequency-selection threshold of +100(1+j) instead of zero threshold (ablation) | RSNet with simple FFT-ReLU stream · GoPro test (Group I) | PSNR: 29.51 -> 29.60 | Helped |
| Select only high- or low-frequency components at 1/8 (ablation) | RSNet with simple FFT-ReLU stream · GoPro test (Group I) | PSNR: 29.51 -> 27.81; 29.51 -> 27.83 | Hurt |
| Keep only low-frequency components inside radius 1/4 H (ablation) | RSNet with FFT-ReLU stream · GoPro test (Group I) | PSNR: 30.30 -> 30.44 | Helped |
Closest papers by meaning
- Beyond Spatial Domain: Cross-domain Promoted Fourier Convolution Helps Single Image Dehazing (AAAI 2025 · similarity 0.89)
- Efficient RAW Image Deblurring with Adaptive Frequency Modulation (NeurIPS 2025 · similarity 0.89)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.87)
- BlurDM: A Blur Diffusion Model for Image Deblurring (NeurIPS 2025 · similarity 0.86)
- Parameterized Blur Kernel Prior Learning for Local Motion Deblurring (CVPR 2025 · similarity 0.86)
- Improving the Learning Capability of Small-size Image Restoration Network by Deep Fourier Shifting (NeurIPS 2024 · 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.