Research map / Image generation / Image restoration
Rawformer: Unpaired Raw-to-Raw Translation for Learnable Camera ISPs
Georgy Perevozchikov, Nancy Mehta, Mahmoud Afifi, Radu Timofte
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 self-supervised inpainting pre-training of the generator (ablation) | Rawformer · NUS Canon-to-Nikon | PSNR: 38.73 -> 41.89 | Improved |
| replace L1 loss with L1+VGG then pixel-wise (L1+SSIM+VGG) loss (ablation) | Rawformer · NUS Canon-to-Nikon | PSNR: 37.13 -> 37.94 -> 38.73 | Improved |
| add condensed query attention (CQA) to baseline UNet generator (ablation) | Rawformer generator · NUS Canon-to-Nikon | PSNR: 37.23 -> 38.21 | Improved |
| add style modulator (ablation) | Rawformer generator · NUS Canon-to-Nikon | PSNR: 38.21 -> 39.08 | Improved |
| add scale perceptive feed-forward network (SPFN) (ablation) | Rawformer generator · NUS Canon-to-Nikon | PSNR: 39.08 -> 39.72 | Improved |
| add composite up/downsampler (CUp-CDown) (ablation) | Rawformer generator · NUS Canon-to-Nikon | PSNR: 39.72 -> 39.93 | Improved |
| add discriminator head with cache on top of attention and batchnorm (ablation) | Rawformer discriminator · NUS Canon-to-Nikon | PSNR: 40.24 -> 41.89 | Improved |
| compress queries with average pooling plus linear projection instead of other projections (ablation) | CQA block · NUS Canon-to-Nikon | PSNR: 38.83 (patch merging) -> 41.89 | Improved |
| use multi-scale instead of single-scale SPFN (ablation) | Rawformer · NUS | PSNR: 41.11 -> 41.89 | Improved |
| int8 post-training quantization for mobile deployment | Rawformer · Raw-to-Raw Samsung-to-iPhone | PSNR: 40.98 -> 37.21 (CPU int8, 526 -> 179 ms) | Hurt |
| convert to fp16 with TensorRT | Rawformer · Raw-to-Raw Samsung-to-iPhone | PSNR / time: 40.98 -> 40.92; 26 -> 18 ms | No clear effect |
Later papers that compared with this method
1 later accepted paper reports Rawformer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- MERIT: Multi-domain Efficient RAW Image Translation (CVPR 2026 · better on 5 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RAW-Adapter: Adapting Pre-Trained Visual Model to Camera RAW Images and a Benchmark (TPAMI 2026 · similarity 0.90)
- Transform Your Smartphone into a DSLR Camera: Learning the ISP in the Wild (ECCV 2022 · similarity 0.89)
- DiffRAW: Leveraging Diffusion Model to Generate DSLR-Comparable Perceptual Quality sRGB from Smartphone RAW Images (AAAI 2024 · similarity 0.84)
- ParamISP: Learned Forward and Inverse ISPs using Camera Parameters (CVPR 2024 · similarity 0.83)
- Compress Clean Signal from Noisy Raw Image: A Self-Supervised Approach (ICML 2024 · similarity 0.83)
- ISPDiffuser: Learning RAW-to-sRGB Mappings with Texture-Aware Diffusion Models and Histogram-Guided Color Consistency (AAAI 2025 · 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.
