Research map / Image generation / Image restoration
SpiralDiff: Spiral Diffusion with LoRA for RGB-to-RAW Conversion Across Cameras
Huanjing Yue, Shangbin Xie, Cong Cao, Qian Wu, Lei Zhang, Lei Zhao, Jingyu Yang
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 |
|---|---|---|---|
| use SpiralDiff instead of RAW-Diffusion (separate) | RGB-to-RAW conversion · FiveK Canon | PSNR (dB): 39.96 -> 42.82 | Improved |
| add time-varying signal-dependent noise weighting w_t (ablation) | ResShift baseline · FiveK Canon | PSNR (dB): 41.40 -> 42.82 | Improved |
| add time-varying signal-dependent noise weighting w_t (ablation) | ResShift baseline · NOD Nikon | PSNR (dB): 53.48 -> 53.64 | No clear effect |
| weight noise by static RGB y0 instead of w_t (ablation) | SpiralDiff · FiveK Canon | PSNR (dB): 41.40 -> 40.06 | Hurt |
| merge cameras in combined training without camera info (ablation) | SpiralDiff · FiveK Canon | PSNR (dB): 42.82 -> 41.99 | Hurt |
| merge cameras in combined training (small FiveK Nikon set) (ablation) | SpiralDiff · FiveK Nikon | PSNR (dB): 41.72 -> 43.19 | Improved |
| add CamLoRA to the combined model (ablation) | SpiralDiff · NOD Nikon | PSNR (dB): 52.06 -> 52.62 | Improved |
| use direct camera embedding instead of CamLoRA (ablation) | SpiralDiff combined · FiveK Canon | PSNR (dB): 41.99 -> 40.97 | Hurt |
| use SpiralDiff diffusion process in RAW-Diffusion | RAW-Diffusion · Real-ISP iPhone | PSNR (dB): 22.74 -> 30.11 | Improved |
| fine-tune CamLoRA with 1 image instead of training from scratch | SpiralDiff · SID Sony 1-shot | PSNR (dB): 39.83 -> 42.85 | Improved |
| add synthesized RAW from BDD100K to 100 real RAW images | RAW detector · NOD Nikon | AP: 18.4 -> 26.7 | Improved |
| train detector on 100 real RAW images instead of RGB | detector · NOD Nikon | AP: 19.1 -> 18.4 | Hurt |
Closest papers by meaning
- ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge (CVPR 2025 · similarity 0.87)
- RawGen: Learning Camera Raw Image Generation (ECCV 2026 · similarity 0.86)
- RawMetaDiff: Unlocking Extreme Darkness from Dual-Exposure RAW with Meta-Guided Diffusion (CVPR 2026 · similarity 0.86)
- DiffRAW: Leveraging Diffusion Model to Generate DSLR-Comparable Perceptual Quality sRGB from Smartphone RAW Images (AAAI 2024 · similarity 0.85)
- RAWild: Toward Sensor-Agnostic RAW Object Detection via Physics-Guided Curve and Grid Modeling (NeurIPS 2026 · similarity 0.84)
- MultiDiff: Consistent Novel View Synthesis from a Single Image (CVPR 2024 · similarity 0.84)
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.
