Research map / Image generation / Image restoration
Self-Supervised Image Restoration with Blurry and Noisy Pairs
Zhilu Zhang, RongJian Xu, Ming Liu, Zifei Yan, Wangmeng Zuo
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 |
|---|---|---|---|
| jointly use the short-exposure noisy and long-exposure blurry images for restoration | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 1.45 dB | Improved |
| jointly use the short-exposure noisy and long-exposure blurry images for restoration | SelfIR · Synthetic sRGB Poisson noise | PSNR: 1.59 dB | Improved |
| jointly use the short-exposure noisy and long-exposure blurry images for restoration | SelfIR · Synthetic raw-RGB sensor noise | PSNR: 1.69 dB | Improved |
| replace clear-image supervision with noisy-image supervision for deblurring (ablation) | Baseline_B · Synthetic deblurring with Gaussian, Poisson, and sensor noise | PSNR: 28.24 -> 28.29; 28.24 -> 28.23; 28.14 -> 28.16 | No clear effect |
| add blurry-image auxiliary supervision to Neighbor2Neighbor (ablation) | Neighbor2Neighbor · Synthetic sRGB Gaussian noise | PSNR: 34.29 -> 34.45 | Improved |
| add auxiliary loss (ablation) | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 35.65 -> 35.74 | Improved |
| increase regularization-loss weight from 0 to 2 (ablation) | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 35.20 -> 35.74 | Improved |
| increase regularization-loss weight from 2 to 8 (ablation) | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 35.74 -> 35.73 | No clear effect |
| increase auxiliary-loss weight from 0 to 2 (ablation) | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 35.65 -> 35.74 | Improved |
| increase auxiliary-loss weight from 2 to 8 (ablation) | SelfIR · Synthetic sRGB Gaussian noise | PSNR: 35.74 -> 35.67 | Hurt |
| fine-tune on the real-world training pairs | SelfIR · Real-world raw-RGB images | PI: 0.55 | Improved |
Closest papers by meaning
- Iterative Denoiser and Noise Estimator for Self-Supervised Image Denoising (ICCV 2023 · similarity 0.89)
- Unsupervised Image Restoration Using Partially Linear Denoisers (TPAMI 2021 · similarity 0.89)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.89)
- Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · similarity 0.88)
- Unsupervised Blind Image Deblurring Based on Self-Enhancement (CVPR 2024 · similarity 0.87)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.87)
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.
