Research map / Image generation / Image restoration
Self-Guided Network for Fast Image Denoising
Shuhang Gu, Yawei Li, Luc Van Gool, Radu Timofte
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 one top-down guidance level (L0 -> L1) (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 26.95 -> 28.03 | Improved |
| add one top-down guidance level (L0 -> L1) (ablation) | SGN · BSD68 gray, sigma=50 | PSNR (dB): 25.48 -> 26.15 | Improved |
| increase levels from L1 to L2 (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 28.03 -> 28.42 | Improved |
| increase levels from L3 to L4 (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 28.57 -> 28.58 | No clear effect |
| increase levels from L3 to L4 (time) (ablation) | SGN · DIV2K 100 gray, sigma=50 | runtime: 0.0942 -> 0.1324 | Hurt |
| introduce contextual guidance early rather than late (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 27.92 -> 28.08 | Improved |
| introduce guidance early rather than middle (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 28.02 -> 28.08 | Improved |
| introduce guidance early rather than late (ablation) | SGN · Set68 gray, sigma=50 | PSNR (dB): 26.06 -> 26.16 | Improved |
| use pixel shuffling instead of down-sampling to make multi-scale inputs (ablation) | SGN · DIV2K 100 gray, sigma=50 | PSNR (dB): 28.18 -> 28.53 | Improved |
| use pixel shuffling instead of down-sampling to make multi-scale inputs (ablation) | SGN · BSD68 gray, sigma=50 | PSNR (dB): 26.20 -> 26.43 | Improved |
| SGN vs U-Net on SID Fuji | SGNL4 · SID Fuji | PSNR (dB): 26.61 -> 27.41 | Improved |
| SGNL3 vs MemNet on DIV2K gray sigma=50 | SGNL3 · DIV2K 100 gray, sigma=50 | PSNR (dB): 28.50 -> 28.53 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MemNet — MemNet: A Persistent Memory Network for Image Restoration (ICCV 2017) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Joint Demosaicing and Denoising With Self Guidance (CVPR 2020 · similarity 0.89)
- LG-BPN: Local and Global Blind-Patch Network for Self-Supervised Real-World Denoising (CVPR 2023 · similarity 0.86)
- Rethinking Transformer-Based Blind-Spot Network for Self-Supervised Image Denoising (AAAI 2025 · similarity 0.86)
- Image Restoration by Denoising Diffusion Models with Iteratively Preconditioned Guidance (CVPR 2024 · similarity 0.86)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.85)
- Exploring Efficient Asymmetric Blind-Spots for Self-Supervised Denoising in Real-World Scenarios (CVPR 2024 · similarity 0.85)
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.
