Research map / Probabilistic and scientific modeling / Molecules and proteins
Physics-Guided ISO-Dependent Sensor Noise Modeling for Extreme Low-Light Photography
Read the paperCheck 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 physics-guided ISO-dependent noise modeling with real noise-based synthesis instead of PMN | Low-light denoising on SID and ELD · Sony A7S2 and Nikon D850 | PSNR: 42.31 -> 43.36 (SID ×100); 40.18 -> 41.02 (SID ×250); 36.97 -> 37.80 (SID ×300); 45.74 -> 46.74 (ELD Sony ×100); 43.84 - | Helped |
| Add black level error noise (NBLE) (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 39.62 -> 41.65 (SID ×100); 35.28 -> 39.56 (SID ×250); 31.98 -> 36.37 (SID ×300); 43.08 -> 44.86 (ELD ×100); 40.02 -> 42. | Helped |
| Add quantization noise (Nq) to NHG and NBLE (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 41.65 -> 40.87; 39.56 -> 37.00; 36.37 -> 33.77; 44.86 -> 44.75; 42.99 -> 42.50 | Mixed |
| Add row noise (Nrow) to NHG and NBLE (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 40.87 -> 41.89; 37.00 -> 39.54; 33.77 -> 36.43; 44.75 -> 45.28; 42.50 -> 43.33 | Helped |
| Add dark-current fixed-pattern noise (k·NF P) (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 41.89 -> 42.10; 39.54 -> 39.76; 36.43 -> 36.76; 45.28 -> 45.61; 43.33 -> 43.84 | Helped |
| Use two-stage random-then-fixed-pattern training instead of one-stage training (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 41.23 -> 42.10; 37.11 -> 39.76; 33.80 -> 36.76; 45.03 -> 45.61; 42.72 -> 43.84 | Helped |
| Model black level error noise as ISO-dependent instead of using ELLE BLE (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 41.78 -> 42.10; 39.66 -> 39.76; 36.65 -> 36.76; 45.22 -> 45.61; 43.34 -> 43.84 | Helped |
| Model black level error noise as ISO-dependent instead of using ELD BLE (ablation) | Proposed sensor noise model · SID and ELD | PSNR: 42.02 -> 42.10; 39.73 -> 39.76; 36.59 -> 36.76; 45.11 -> 45.61; 43.32 -> 43.84 | Helped |
| Train only on paired noisy-clean images (D, I) rather than the proposed combined real-noise synthesis strategies (ablation) | Denoising network using the proposed noise model · SID and ELD | PSNR: 43.36 vs 42.06 (SID ×100); 41.02 vs 39.60 (SID ×250); 37.80 vs 36.85 (SID ×300); 46.74 vs 44.47 (ELD ×100); 44.95 vs 41. | Hurt |
| Use darkshading-corrected real pairs without zero-mean synthetic noise pairs (ablation) | Denoising network using the proposed noise model · SID and ELD | PSNR: 43.36 vs 42.84 (SID ×100); 41.02 vs 40.53 (SID ×250); 37.80 vs 37.57 (SID ×300); 46.74 vs 46.12 (ELD ×100); 44.95 vs 44. | Hurt |
| Use zero-mean synthetic noise pairs without darkshading-corrected real pairs (ablation) | Denoising network using the proposed noise model · SID and ELD | PSNR: 43.36 vs 42.75 (SID ×100); 41.02 vs 40.41 (SID ×250); 37.80 vs 37.15 (SID ×300); 46.74 vs 46.24 (ELD ×100); 44.95 vs 44. | Mixed |
Closest papers by meaning
- A Physics-Based Noise Formation Model for Extreme Low-Light Raw Denoising (CVPR 2020 · similarity 0.91)
- Noise2NoiseFlow: Realistic Camera Noise Modeling Without Clean Images (CVPR 2022 · similarity 0.87)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.84)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · similarity 0.84)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.84)
- LED: A Large-scale Real-world Paired Dataset for Event Camera Denoising (CVPR 2024 · similarity 0.83)
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.