Research map / Image generation / Image restoration
MambaRaw: Selective State Space Modeling for Efficient 4K RAW Image Reconstruction
Peize Li, Fanhu Zeng, Tongda Xu, XINJIE ZHANG, Xingtong Ge, Haotian Zhang, Xingguo Xu, Yan 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 |
|---|---|---|---|
| add Energy-Aware Refinement (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 58.21 -> 58.55 | Improved |
| add dense SSM context model (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 58.55 -> 59.61 | Improved |
| add tile-wise selection to dense SSM (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | Time (ms): 584 -> 515 | Improved |
| add tile-wise selection to dense SSM (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 59.61 -> 59.58 | No clear effect |
| select tiles by L2 energy instead of random (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 59.25 -> 59.58 | Improved |
| select tiles by L2 energy instead of entropy (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | Time (ms): 542 -> 515 | Improved |
| use SSM instead of CNN block (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 58.45 -> 59.58 | Improved |
| use SSM instead of window-attention transformer (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | Time (ms): 620 -> 515 | Improved |
| raise keep ratio from 0.5 to 1.0 (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 59.58 -> 59.61 | No clear effect |
| lower keep ratio from 0.5 to 0.25 (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 59.58 -> 59.42 | Hurt |
| use tile size 128 instead of 64 (ablation) | MambaRaw · Sony SLT-A57 (lambda 0.8) | PSNR: 59.58 -> 59.50 | Hurt |
| MambaRaw vs Beyond-R2LCM at true 4K | MambaRaw · True 4K | FLOPs / PSNR: 5420.7 -> 2380.5 GFLOPs; 55.21 -> 56.58 | Improved |
Closest papers by meaning
- MambaIRv2: Attentive State Space Restoration (CVPR 2025 · similarity 0.86)
- LC-Mamba: Local and Continuous Mamba with Shifted Windows for Frame Interpolation (CVPR 2025 · similarity 0.85)
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.85)
- MaIR: A Locality- and Continuity-Preserving Mamba for Image Restoration (CVPR 2025 · similarity 0.85)
- MambaIC: State Space Models for High-Performance Learned Image Compression (CVPR 2025 · similarity 0.85)
- MambaSCI: Efficient Mamba-UNet for Quad-Bayer Patterned Video Snapshot Compressive Imaging (NeurIPS 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.
