Research map / Image generation / Image restoration
Resolving Blind Inverse Problems under Dynamic Range Compression via Structured Forward Operator Modeling
Muyu Liu, Xuanyu Tian, Chenhe Du, Qing Wu, Hongjiang Wei, Yuyao Zhang
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 cascade of low-degree monotonic Bernstein polynomials (N=3, K=8) instead of one degree-64 polynomial (ablation) | CaMB-Diff · LOLv1 | PSNR: 10.99 -> 20.14 | Improved |
| raise per-stage degree from 2 to 3 in cascade (K=8) (ablation) | CaMB-Diff · LOLv1 | FID: 87.91 -> 72.65 | Improved |
| replace GDP affine or TAO CNN forward operator with CaMB (16 params) (ablation) | CaMB-Diff · LOLv1 | PSNR: 17.06 -> 20.14 (vs GDP); 17.18 -> 20.14 (vs TAO) | Improved |
| enforce monotonicity as hard architectural constraint in forward operator (ablation) | CaMB-Diff · Low-field MRI operator estimation | validation error | Improved |
| CaMB-Diff versus zero-shot baselines on low-light enhancement | CaMB-Diff · LOLv1, LOLv2 | LOE: LOLv1 LOE 82.21 vs 190.22 (GDP) | Improved |
| CaMB-Diff versus AGLLDiff on low-light PSNR/SSIM | CaMB-Diff · LOLv1, LOLv2 Real, LOLv2 Synthetic | SSIM: LOLv1 SSIM 0.8498 vs 0.8699 (AGLLDiff); LOLv2 Synthetic 0.8748 vs 0.8735 | Mixed |
| CaMB-Diff zero-shot versus supervised HVI under domain shift | CaMB-Diff · LOLv1 OOD (trained on LOLv2-synthetic) | PSNR: 13.63 -> 20.14 | Improved |
| CaMB-Diff versus supervised methods in-domain | CaMB-Diff · LOLv1 | PSNR: 20.14 vs 26.20 (Retinexformer) | Hurt |
| CaMB-Diff vs zero-shot baselines on low-field MRI | CaMB-Diff · HCP T1w | PSNR: 23.61 -> 26.43 (vs TAO) | Improved |
| CaMB-Diff vs GDP/TAO on HDR reconstruction | CaMB-Diff · ImageNet synthetic HDR | PSNR: 16.80 -> 19.95 | Improved |
| CaMB-Diff vs GDP/TAO on real HDR-Eye | CaMB-Diff · HDR-Eye | PSNR: 15.35 -> 17.78 | Improved |
| use U-shaped guidance schedule instead of standard HQS schedule (ablation) | CaMB-Diff · zero-shot UDRC | qualitative convergence | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AGLLDiff — AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025) · better on 2 of 2 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
- Generating Content for HDR Deghosting from Frequency View (CVPR 2024 · similarity 0.85)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.84)
- NPN: Non-Linear Projections of the Null-Space for Imaging Inverse Problems (NeurIPS 2025 · similarity 0.84)
- Single Shot HDR Recovery via a Video Diffusion Prior (NeurIPS 2026 · similarity 0.84)
- Blind Image Restoration via Fast Diffusion Inversion (NeurIPS 2024 · similarity 0.84)
- A Diffusion Model with State Estimation for Degradation-Blind Inverse Imaging (AAAI 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.
