Research map / Image generation / Fast diffusion sampling
Kernel Diffusion: An Alternate Approach to Blind Deconvolution
Yash Sanghvi, Yiheng Chi, Stanley Chan
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 non-blind solver (DWDN) reblurring-loss guidance to kernel diffusion (ablation) | Kernel-Diff · Ablation set | FID: 152.51 -> 140.28 | Improved |
| use guidance step size rho 0.05 instead of 0.5 (ablation) | Kernel-Diff · Ablation set | PSNR: 20.87 -> 22.64 | Improved |
| use guidance step size rho 0.05 instead of 0.1 (ablation) | Kernel-Diff · Ablation set | MNC: 0.567 -> 0.724 | Improved |
| add Stage II kernel refinement (ablation) | Kernel-Diff · Ablation set | PSNR: 21.76 -> 22.64 | Improved |
| estimate the kernel with diffusion (kernel-first) instead of alternating image and kernel diffusion (Blind-DPS) | blind deconvolution · BSD100 synthetic blur | FID: 280.53 -> 172.33 | Improved |
| kernel-first estimation instead of alternating minimization | blind deconvolution · BSD100 kernel accuracy | MNC: 0.283 -> 0.663 | Improved |
| kernel-first diffusion with lightweight model instead of Blind-DPS | blind deconvolution · 256x256 image | parameters / time: 110M -> 43M parameters; 337s -> 305s | Improved |
| kernel-first estimation (Kernel-Diff) versus alternating minimization | blind deconvolution · reblurring residual across iterations | residual | Improved |
Closest papers by meaning
- Blind Image Deconvolution by Automatic Gradient Activation (CVPR 2016 · similarity 0.91)
- Normalized Blind Deconvolution (ECCV 2018 · similarity 0.91)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Blind Image Restoration via Fast Diffusion Inversion (NeurIPS 2024 · similarity 0.88)
- A Diffusion Model with State Estimation for Degradation-Blind Inverse Imaging (AAAI 2024 · similarity 0.88)
- Taming Generative Diffusion Prior for Universal Blind Image Restoration (NeurIPS 2024 · similarity 0.88)
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.
