Research map / Image generation / Image restoration
Noise2NoiseFlow: Realistic Camera Noise Modeling Without Clean Images
Check 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 |
|---|---|---|---|
| Jointly train the denoiser and Noise Flow noise model using cross-sample noisy-image pairs instead of clean/noisy pairs | Noise2NoiseFlow · SIDD test | D_KL: 0.0267 -> 0.0265 | Helped |
| Jointly train the denoiser and noise model rather than train Noise2Noise denoiser and Noise Flow separately | Noise2NoiseFlow · SIDD test | D_KL: 0.0363 -> 0.0265 | Helped |
| Jointly train noise model and denoiser on noisy/noisy pairs | Noise2NoiseFlow · SIDD-Medium test | PSNR: 51.57 -> 52.80 | Helped |
| Use joint noisy-pair training instead of supervised clean/noisy DnCNN training | Noise2NoiseFlow · SIDD-Medium test | PSNR: 51.32 -> 52.80 | Helped |
| Use U-Net instead of DnCNN as the denoiser (ablation) | Noise2NoiseFlow · SIDD test | PSNR: 52.80 -> 52.64 | No clear effect |
| Replace Noise Flow with NLF noise model (ablation) | Noise2NoiseFlow · SIDD test | D_KL: 0.0265 -> 0.0288 | Hurt |
| Replace Noise Flow with NLF noise model (ablation) | Noise2NoiseFlow · SIDD test | PSNR: 52.80 -> 52.69 | No clear effect |
| Set Noise2Noise loss weight to zero (ablation) | Noise2NoiseFlow · SIDD test | NLL | No clear effect |
| Use self-sample loss instead of cross-sample loss (ablation) | Noise2NoiseFlow · SIDD test | NLL | Hurt |
| Train the noise model and denoiser with individual noisy images using R2R augmentation rather than paired noisy observations (ablation) | Noise2NoiseFlow · SIDD test | D_KL: 0.0265 -> 0.0983 | Hurt |
Closest papers by meaning
- Estimating Fine-Grained Noise Model via Contrastive Learning (CVPR 2022 · similarity 0.91)
- Convexity-Aware Noise Calibration: A Self-Supervised Framework for Noise-Level-Unknown Image Denoising (CVPR 2026 · similarity 0.90)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.89)
- NERDS: A General Framework to Train Camera Denoisers from Raw-RGB Noisy Image Pairs (ICLR 2023 · similarity 0.89)
- Noisier2Noise: Learning to Denoise From Unpaired Noisy Data (CVPR 2020 · similarity 0.89)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.88)
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.