Research map / Image generation / Image restoration
Unprocessing Images for Learned Raw Denoising
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 |
|---|---|---|---|
| train with unprocessed synthetic raw data and a processing-aware loss | raw denoising on the Darmstadt Noise Dataset · Darmstadt Noise Dataset, raw | PSNR: 47.56 -> 48.89 | Helped |
| train with unprocessed synthetic raw data and a processing-aware loss | raw denoising on the Darmstadt Noise Dataset · Darmstadt Noise Dataset, raw | SSIM: 0.9767 -> 0.9824 | Helped |
| train with unprocessed synthetic raw data and a processing-aware loss | sRGB denoising on the Darmstadt Noise Dataset · Darmstadt Noise Dataset, sRGB | PSNR: 38.32 -> 40.35 | Helped |
| train with unprocessed synthetic raw data and a processing-aware loss | sRGB denoising on the Darmstadt Noise Dataset · Darmstadt Noise Dataset, sRGB | SSIM: 0.9421 -> 0.9641 | Helped |
| use the sRGB-targeted rather than raw-targeted training loss | the proposed denoising model · Darmstadt Noise Dataset, sRGB | PSNR: 40.17 -> 40.35 | Helped |
| use the raw-targeted rather than sRGB-targeted training loss | the proposed denoising model · Darmstadt Noise Dataset, raw | PSNR: 48.88 -> 48.89 | No clear effect |
| use the proposed model rather than N3Net | Darmstadt Noise Dataset denoising · Darmstadt Noise Dataset | runtime (ms): 210 -> 22 | Helped |
| use the proposed model rather than CBDNet | Darmstadt Noise Dataset denoising · Darmstadt Noise Dataset | runtime (ms): 400 -> 22 | Helped |
| use realistic shot-and-read noise and provide the noise level to the network (ablation) | Our Model (sRGB) · Darmstadt Noise Dataset | sRGB PSNR: 38.65 -> 40.35 | Helped |
| add unprocessing steps to synthesize raw training data (ablation) | Our Model (sRGB) · Darmstadt Noise Dataset | sRGB PSNR: 39.02 -> 40.35 | Helped |
| add unprocessing while keeping the network at its original size (ablation) | Our Model (sRGB), compared with the larger no-unprocessing variant · Darmstadt Noise Dataset | sRGB PSNR: 39.35 -> 40.35 | Helped |
| invert CCM, white balance, and digital gain during synthesis (ablation) | Our Model (sRGB) · Darmstadt Noise Dataset | sRGB PSNR: 39.70 -> 40.35 | Helped |
Closest papers by meaning
- CycleISP: Real Image Restoration via Improved Data Synthesis (CVPR 2020 · similarity 0.90)
- Model-Based Image Signal Processors via Learnable Dictionaries (AAAI 2022 · similarity 0.89)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.89)
- Toward Convolutional Blind Denoising of Real Photographs (CVPR 2019 · similarity 0.89)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.87)
- Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable (CVPR 2025 · similarity 0.87)
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.