Research map / Probabilistic and scientific modeling / Variational and generative sampling
Noise Flow: Noise Modeling With Conditional Normalizing Flows
Abdelrahman Abdelhamed, Marcus A. Brubaker, Michael S. Brown
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 camera-specific gain parameters to the signal-dependent and gain layers (ablation) | Noise Flow (S-G) · SIDD noise modeling | NLL: -3.431 -> -3.511 | Improved |
| add camera-specific gain parameters (ablation) | Noise Flow (S-G) · SIDD noise synthesis | marginal KL divergence: 0.067 -> 0.010 | Improved |
| add one unconditional flow step (affine coupling + 1x1 conv) around each layer (ablation) | Noise Flow (S-G-CAM) · SIDD noise modeling | NLL: -3.511 -> -3.518 | Improved |
| increase unconditional flow steps from one to four (ablation) | Noise Flow · SIDD noise modeling | NLL: -3.518 -> -3.521 | Improved |
| increase unconditional flow steps from one to four (ablation) | Noise Flow · SIDD noise synthesis | marginal KL divergence: 0.009 -> 0.008 | Improved |
| replace Gaussian model with Noise Flow | noise density model · SIDD noise modeling | NLL: -2.831 -> -3.521 | Improved |
| replace camera NLF with Noise Flow | noise density model · SIDD noise modeling | NLL: -3.105 -> -3.521 | Improved |
| replace camera NLF with Noise Flow for noise synthesis | noise synthesis · SIDD noise synthesis | marginal KL divergence: 0.052 -> 0.008 | Improved |
| train DnCNN on Noise Flow noise instead of Gaussian noise | DnCNN · SIDD-Validation denoising | PSNR: 43.63 -> 48.52 | Improved |
| train DnCNN on Noise Flow noise instead of camera NLF noise | DnCNN · SIDD-Validation denoising | PSNR: 44.99 -> 48.52 | Improved |
| train DnCNN on Noise Flow synthetic noise instead of real noisy/clean pairs | DnCNN · SIDD-Validation denoising | PSNR: 47.08 -> 48.52 | Improved |
| train DnCNN on Noise Flow noise instead of real noise | DnCNN · SIDD-Validation denoising | SSIM: 0.989 -> 0.992 | Improved |
Closest papers by meaning
- Noise2NoiseFlow: Realistic Camera Noise Modeling Without Clean Images (CVPR 2022 · similarity 0.91)
- Modeling sRGB Camera Noise With Normalizing Flows (CVPR 2022 · similarity 0.90)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.88)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.86)
- Asymmetric Flow Models (NeurIPS 2026 · similarity 0.85)
- Normalizing Trajectory Models (NeurIPS 2026 · 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.
