Research map / Image generation / Image restoration
Unpaired Learning of Deep Image Denoising
Xiaohe Wu, Ming Liu, Yue Cao, Dongwei Ren, 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 |
|---|---|---|---|
| add knowledge distillation to MWCNN with unpaired clean images on top of D-BSN (ablation) | D-BSN · BSD68 gray AWGN sigma=25 | PSNR (dB): 29.12 -> 29.38 | Improved |
| add knowledge distillation to MWCNN with unpaired clean images on top of D-BSN (ablation) | D-BSN · BSD68 gray heteroscedastic Gaussian | PSNR (dB): 29.16 -> 30.10 | Improved |
| add knowledge distillation to MWCNN with unpaired clean images on top of D-BSN (ablation) | D-BSN · CBSD68 color AWGN sigma=25 | PSNR (dB): 30.61 -> 31.40 | Improved |
| add knowledge distillation to MWCNN with unpaired clean images on top of D-BSN (ablation) | D-BSN · CBSD68 color heteroscedastic Gaussian | PSNR (dB): 30.56 -> 31.72 | Improved |
| add knowledge distillation to MWCNN with unpaired clean images on top of D-BSN (ablation) | D-BSN · CBSD68 color multivariate Gaussian | PSNR (dB): 26.48 -> 26.81 | Improved |
| train MWCNN with unpaired data instead of noisy-noisy pairs (N2N) | MWCNN (N2N) · CBSD68 color heteroscedastic Gaussian | PSNR (dB): 31.13 -> 31.72 | Improved |
| train MWCNN with unpaired data instead of paired supervision (N2C) | MWCNN (N2C) · BSD68 gray AWGN sigma=25 | PSNR (dB): 29.41 -> 29.38 | No clear effect |
| train MWCNN with unpaired data instead of paired supervision (N2C) | MWCNN (N2C) · CBSD68 color heteroscedastic Gaussian | PSNR (dB): 32.10 -> 31.72 | Hurt |
| use pixel-shuffle downsampling with factor 4 for real noisy photographs (ablation) | D-BSN · real-world noisy photographs | PSNR | Hurt |
| use MWCNN (unpaired) instead of GCBD | GCBD · DND sRGB real noise | PSNR (dB): 35.58 -> 37.93 | Improved |
| use MWCNN (unpaired) instead of N2S/N2V | N2V · CC15 real noise | PSNR (dB): 35.27 -> 35.90 | Improved |
| use MWCNN (unpaired) instead of CBDNet | CBDNet · CBSD68 color heteroscedastic Gaussian | PSNR (dB): 30.89 -> 31.72 | Improved |
Closest papers by meaning
- End-to-End Unpaired Image Denoising with Conditional Adversarial Networks (AAAI 2020 · similarity 0.93)
- High-Quality Self-Supervised Deep Image Denoising (NeurIPS 2019 · similarity 0.93)
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.93)
- Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks (ICLR 2020 · similarity 0.92)
- Unsupervised Image Restoration Using Partially Linear Denoisers (TPAMI 2021 · similarity 0.92)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.90)
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.
