Research map / Image generation / Image restoration
Modeling sRGB Camera Noise With Normalizing Flows
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 |
|---|---|---|---|
| Use the proposed conditional normalizing flow instead of Noise Flow | sRGB noise modeling · SIDD test set | NLL: 3.311 -> 3.072 | Helped |
| Use the proposed conditional normalizing flow instead of the best baseline, Diagonal Gaussian | sRGB noise modeling · SIDD test set | D_KL: 0.079 -> 0.044 | Helped |
| Train DnCNN using noise synthesized by the proposed model instead of Noise Flow | DnCNN · SIDD-Benchmark | PSNR: 33.81 -> 34.74 | Helped |
| Train DnCNN using noise synthesized by the proposed model instead of C2N | DnCNN · SIDD-Benchmark | SSIM: 0.901 -> 0.912 | Helped |
| Condition CCS on all important variables rather than only ISO (ablation) | sRGB noise model · SIDD test set | NLL: 3.726 -> 3.072 | Helped |
| Condition CCS on all important variables rather than only camera (ablation) | sRGB noise model · SIDD test set | D_KL: 0.216 -> 0.044 | Helped |
| Add a conditional linear flow to conditional coupling steps (ablation) | sRGB noise model · SIDD test set | NLL: 3.530 -> 3.072 | Helped |
| Increase the number of CL-CCS flow blocks from one to four (ablation) | sRGB noise model · SIDD test set | D_KL: 0.075 -> 0.044 | Helped |
| Add an inverse gamma transformation (ablation) | Our model · SIDD test set | D_KL: 0.044 -> 0.061 | Hurt |
| Remove unconditional spline coupling layers (ablation) | one-block spline-based flow · SIDD test set | NLL: 4.238 -> 3.076 | Helped |
| Remove unconditional affine coupling layers (ablation) | one-block affine-based flow · SIDD test set | NLL: 4.147 -> 3.602 | Helped |
| Add a specialized clean-image-only layer (ablation) | one-block flow with CL and CACx2 · SIDD test set | D_KL: 0.077 -> 0.104 | Hurt |
Closest papers by meaning
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.90)
- Diffusion-Based sRGB Real Noise Generation via Prompt-Driven Noise Representation Learning (CVPR 2026 · similarity 0.90)
- DiffRAW: Leveraging Diffusion Model to Generate DSLR-Comparable Perceptual Quality sRGB from Smartphone RAW Images (AAAI 2024 · similarity 0.85)
- Real-World Mobile Image Denoising Dataset with Efficient Baselines (CVPR 2024 · similarity 0.85)
- ISPDiffuser: Learning RAW-to-sRGB Mappings with Texture-Aware Diffusion Models and Histogram-Guided Color Consistency (AAAI 2025 · similarity 0.85)
- 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.