Research map / Image generation / Image restoration
GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images
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 |
|---|---|---|---|
| Vary patch-selection threshold λ (ablation) | G2G2 · Gaussian, mixture, and correlated noise | PSNR / SSIM: Gaussian: 28.93/0.8293, 28.84/0.8045, 28.80/0.8060; mixture: 40.30/0.9845, 40.16/0.9838, 40.28/0.9848; correlated: 27.85 | Mixed |
| Pair synthesized noisy samples with previous denoiser output rather than a second synthesized noisy sample (ablation) | G2G3 · Gaussian, mixture, correlated, WF and Medical | PSNR / SSIM: 29.02 / 0.8153 vs 28.96 / 0.8080; 30.70 / 0.8621 vs 30.49 / 0.8538; 24.04 / 0.8673 vs 28.00 / 0.8447; 16.24 / 0.4490 vs | Mixed |
| Remove cycle loss L_cyc (ablation) | GAN2GAN generative model · Synthetic noise | PSNR / SSIM | No clear effect |
| Remove cycle loss L_cyc (ablation) | GAN2GAN generative model · WF (Avg = 1) | PSNR / SSIM | Hurt |
| Use outer-minimization weights (α,β,γ)=(1,1,1) instead of (5,1,10) (ablation) | GAN2GAN generative model · BSD68 (σ = 25) | denoising performance | Hurt |
| Remove sigmoid activation from gθ2 output (ablation) | GAN2GAN generative model · Training | trainability | Hurt |
Closest papers by meaning
- Image Blind Denoising With Generative Adversarial Network Based Noise Modeling (CVPR 2018 · similarity 0.93)
- End-to-End Unpaired Image Denoising with Conditional Adversarial Networks (AAAI 2020 · similarity 0.92)
- Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks (ICLR 2020 · similarity 0.91)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.90)
- Is Noise Conditioning Necessary for Denoising Generative Models? (ICML 2025 · similarity 0.90)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.89)
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.