Research map / Image generation / Fast diffusion sampling
C2N: Practical Generative Noise Modeling for Real-World Denoising
Geonwoon Jang, Wooseok Lee, Sanghyun Son, Kyoung Mu Lee
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 signal-dependent 1x1 module to the signal-independent generator (ablation) | C2N generator · CBSD68 Poisson noise | PSNR (dB): 35.21 -> 35.80 | Improved |
| add 3x3 convolution modules to the 1x1 generator (ablation) | C2N generator · CBSD68 spatially correlated Gaussian | PSNR (dB): 28.91 -> 31.03 | Improved |
| add signal-dependent 1x1 module to the signal-independent generator (ablation) | C2N generator · CBSD68 spatially correlated Gaussian | PSNR (dB): 26.09 -> 28.91 | Improved |
| add signal-dependent 1x1 module (ablation) | C2N generator · CBSD68 Gaussian sigma=25 | PSNR (dB): 30.69 -> 30.66 | No clear effect |
| add 3x3 convolution modules to the 1x1 generator (ablation) | C2N generator · CBSD68 Gaussian sigma=25 | PSNR (dB): 30.66 -> 30.40 | Hurt |
| add 3x3 convolution modules to the 1x1 generator (ablation) | C2N generator · CBSD68 Poisson noise | PSNR (dB): 35.80 -> 35.23 | Hurt |
| reduce clean samples for the noise generator from 36K to 360 | C2N · SIDD val (clean source S) | PSNR (dB): 34.08 -> 31.84 | Hurt |
| reduce clean samples for the noise generator from 36K to 720 | C2N · SIDD val (clean source S) | PSNR (dB): 34.08 -> 31.98 | Hurt |
| use DIV2K clean images instead of SIDD for denoiser training | C2N · SIDD val (36K) | PSNR (dB): 34.08 -> 31.72 | Hurt |
| use Urban100 clean images instead of SIDD for denoiser training | C2N · SIDD val (36K) | PSNR (dB): 34.08 -> 31.32 | Hurt |
| reduce clean samples from 720 to 360 | C2N · SIDD val (clean source S) | PSNR (dB): 31.98 -> 31.84 | No clear effect |
Closest papers by meaning
- GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images (ICLR 2021 · similarity 0.87)
- Noisier2Noise: Learning to Denoise From Unpaired Noisy Data (CVPR 2020 · similarity 0.87)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.86)
- Is Noise Conditioning Necessary for Denoising Generative Models? (ICML 2025 · similarity 0.86)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · similarity 0.84)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.84)
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.
