Research map / Image generation / Image restoration
Rethinking Noise Synthesis and Modeling in Raw Denoising
Yi Zhang, Hongwei Qin, Xiaogang Wang, Hongsheng Li
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 pattern-aligned patch sampling instead of pixel-wise sampling (ablation) | real-noise sampling · ELD SonyA7S2 x100 | PSNR: 45.71 -> 47.13 | Improved |
| add pattern-aligned patch sampling instead of pixel-wise sampling (ablation) | real-noise sampling · ELD SonyA7S2 x200 | PSNR: 42.57 -> 44.36 | Improved |
| add pattern-aligned patch sampling (Bayer aligned) instead of plain patch sampling (ablation) | real-noise sampling · ELD SonyA7S2 x200 | PSNR: 42.70 -> 44.36 | Improved |
| use plain patch sampling instead of pixel-wise sampling (ablation) | real-noise sampling · ELD SonyA7S2 x100 | PSNR: 45.71 -> 45.83 | No clear effect |
| add high-bit reconstruction to PAP (ablation) | real-noise sampling · ELD SonyA7S2 x100 | PSNR: 47.13 -> 47.90 | Improved |
| add high-bit reconstruction to PAP (ablation) | real-noise sampling · ELD SonyA7S2 x200 | PSNR: 44.36 -> 45.47 | Improved |
| quantize synthetic signal-independent noise to 14-bit in ELD model (low-bit) (ablation) | ELD noise model · ELD SonyA7S2 x200 | PSNR: 45.46 -> 42.35 | Hurt |
| quantize synthetic signal-independent noise to 14-bit in P-G model (ablation) | Poisson-Gaussian model · ELD SonyA7S2 x200 | PSNR: 43.84 -> 42.98 | Hurt |
| quantize noise to 14-bit at normal light (x1) (ablation) | ELD noise model · ELD SonyA7S2 x1 | PSNR: 53.43 -> 53.41 | No clear effect |
| use calibrated P-G parameters instead of in-camera noise profile (ablation) | Poisson-Gaussian model · SIDD Galaxy S6 | PSNR: 34.64 -> 45.38 | Improved |
| use calibrated P-G instead of DNN-based Noise Flow | noise synthesis · SIDD Google Pixel | PSNR: 45.15 -> 46.48 | Improved |
| add high-bit pattern-aligned sampling instead of pixel-wise sampling on SIDD (ablation) | real-noise sampling · SIDD Galaxy S6 | PSNR: 45.33 -> 45.62 | Improved |
Closest papers by meaning
- Calibration-Free Raw Image Denoising via Fine-Grained Noise Estimation (TPAMI 2025 · similarity 0.93)
- Towards General Low-Light Raw Noise Synthesis and Modeling (ICCV 2023 · similarity 0.91)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.91)
- A Physics-Based Noise Formation Model for Extreme Low-Light Raw Denoising (CVPR 2020 · similarity 0.91)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · similarity 0.90)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · 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.
