Research map / 3D vision / Depth and stereo
RAW-Flow: Advancing RGB-to-RAW Image Reconstruction with Deterministic Latent Flow Matching
Zhen Liu, Diedong Feng, Hai Jiang, Liaoyuan Zeng, Hao Wang, Chaoyu Feng, Lei Lei, Bing Zeng, Shuaicheng Liu
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 dual-domain feature alignment loss L_fea to the RAW autoencoder (ablation) | DLAE · RAW autoencoder reconstruction | PSNR: 25.36 -> 32.94 | Improved |
| inject RGB encoder features into the RAW autoencoder (ablation) | DLAE · RAW autoencoder reconstruction | PSNR: 20.08 -> 25.36 | Improved |
| inject RGB features without alignment loss, final RAW reconstruction (ablation) | RAW-Flow · FiveK-Nikon reconstructed RAW | PSNR: 29.19 -> 28.48 | Hurt |
| use deterministic latent flow matching instead of standard diffusion (ablation) | RAW-Flow · FiveK-Nikon | PSNR: 28.30 -> 30.79 | Improved |
| start flow from RGB latent instead of Gaussian noise (ablation) | RAW-Flow · FiveK-Canon | PSNR: 22.74 -> 32.55 | Improved |
| use cross-scale context guidance instead of single-scale context (ablation) | RAW-Flow · FiveK-Canon | PSNR: 28.43 -> 32.55 | Improved |
| use cross-scale context guidance instead of RGB latent guidance (ablation) | RAW-Flow · FiveK-Nikon | PSNR: 28.50 -> 30.79 | Improved |
| flow matching vs best regression baseline (ReRAW-S) | FiveK-Nikon · FiveK-Nikon | PSNR (RAW): 28.04 -> 30.79 | Improved |
| RAW-Flow vs ReRAW-S on PASCALRAW SSIM | PASCALRAW · PASCALRAW | SSIM (RAW): 0.9854 -> 0.9831 | Hurt |
| diffusion-based methods vs best regression method | RGB-to-RAW baselines · FiveK-Nikon | PSNR (RAW): 26.96 vs 28.04 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports RAW-Flow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Bridging RGB and RAW: Single-step Deterministic Flow with Homogeneous Representation Alignment (ICML 2026 · better on 3 of 3 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
- Bridging RGB and RAW: Single-step Deterministic Flow with Homogeneous Representation Alignment (ICML 2026 · similarity 0.91)
- Reconstructing High Quality Raw Video Using Temporal Affinity and Diffusion Prior (TPAMI 2025 · similarity 0.89)
- ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge (CVPR 2025 · similarity 0.86)
- Cross Flow: One-Step Generation Across Latent and Pixel Spaces (NeurIPS 2026 · similarity 0.86)
- ISPDiffuser: Learning RAW-to-sRGB Mappings with Texture-Aware Diffusion Models and Histogram-Guided Color Consistency (AAAI 2025 · similarity 0.86)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · 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.
