Research map / Image generation / Image restoration
Practical Deep Raw Image Denoising on Mobile Devices
Yuzhi Wang, Haibin Huang, Qin Xu, Jiaming Liu, Yiqun Liu, Jue Wang
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 |
|---|---|---|---|
| train on SID-based synthetic data matching sensor noise instead of SIDD | denoising network · Real phone test set | PSNR | Improved |
| use mobile-friendly network instead of UNet-36G | denoiser · Snapdragon 855 GPU | ms per megapixel: 383.9 -> 70.70 | Improved |
| use mobile-friendly network instead of UNet-21G | denoiser · Snapdragon 855 GPU | ms per megapixel: 292.0 -> 70.70 | Improved |
| use accurate noise parameters instead of scaling by 0.5 (ablation) | synthetic noise training · Real phone test set | PSNR (dB): 36.87384014 -> 39.96560621 | Improved |
| use accurate noise parameters instead of scaling by 2.0 (ablation) | synthetic noise training · Real phone test set | PSNR (dB): 39.68042145 -> 39.96560621 | Improved |
| train on SID synthesized data instead of test-set overfitting (ablation) | denoising network · Real phone test set | PSNR (dB): 40.06 -> 39.76 | Mixed |
| add k-Sigma Transform instead of iso-augmentation (ablation) | denoising network · Multi-ISO test set | PSNR (dB): 39.56662978 -> 39.76 | Improved |
| add k-Sigma Transform instead of concat-variance (ablation) | denoising network · Multi-ISO test set | PSNR (dB): 39.653042266876 -> 39.76 | Improved |
| add explicit noise variance channels instead of blind iso-augmentation (ablation) | denoising network · Multi-ISO test set | PSNR (dB): about 0.09 dB higher | Improved |
| train one multi-ISO model with k-Sigma Transform instead of single-ISO 1600 model (ablation) | denoising network · Multi-ISO test set | PSNR (dB): 35.73569171524029 -> 39.76 | Improved |
| train one model with k-Sigma Transform instead of single-ISO model matched to test ISO (ablation) | denoising network · ISO-1600 test images | PSNR (dB): 41.48 -> 41.48 | No clear effect |
| use k-Sigma Transform instead of single-ISO 1600 model at high ISO (ablation) | denoising network · ISO-6400 test images | PSNR (dB): 28.86 -> 36.94 | Improved |
Closest papers by meaning
- Real-World Mobile Image Denoising Dataset with Efficient Baselines (CVPR 2024 · similarity 0.91)
- A High-Quality Denoising Dataset for Smartphone Cameras (CVPR 2018 · similarity 0.91)
- Efficient Real-Time Raw-to-Raw Denoising for Extreme Low-Light Ultra HD Video on Mobile Devices (CVPR 2026 · similarity 0.88)
- Efficient Burst Raw Denoising with Variance Stabilization and Multi-frequency Denoising Network (IJCV 2022 · similarity 0.88)
- NERDS: A General Framework to Train Camera Denoisers from Raw-RGB Noisy Image Pairs (ICLR 2023 · similarity 0.87)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.85)
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.
