Research map / Image generation / Image restoration
Unsupervised Imaging Inverse Problems with Diffusion Distribution Matching
Giacomo Meanti, Thomas Ryckeboer, Michael Arbel, Julien Mairal
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 |
|---|---|---|---|
| learn the blur kernel from a dataset by distribution matching instead of from a single image (Blind-DPS) | DiffPIR deblurring · FFHQ Gaussian blur | PSNR: 29.8 -> 32.7 | Improved |
| learn the blur kernel by distribution matching instead of Blind-DPS | DiffPIR deblurring · FFHQ motion blur | PSNR: 25.5 -> 28.8 | Improved |
| use learned kernel instead of true kernel | DiffPIR · FFHQ motion blur | PSNR: 29.3 -> 28.8 | Hurt |
| use learned degradation instead of DeFlow | ESRGAN · DPDD spatially varying blur | PSNR: 27.92 -> 32.38 | Improved |
| use learned degradation instead of true kernel | DPIR · DPDD spatially varying blur | PSNR: 34.89 -> 33.84 | Hurt |
| use learned degradation instead of true degradation | ESRGAN · DPDD spatially varying blur | LPIPS: 0.058 -> 0.061 | No clear effect |
| use matched-degradation method instead of pretrained Restormer | DPDD deblurring · DPDD spatially varying blur | PSNR: 28.59 -> 33.84 | Improved |
| use learned kernel with USRNet instead of KernelGAN estimate | x2 super-resolution · DIV2KRK | Y-PSNR: 29.65 -> 31.11 | Improved |
| plug learned kernel into DANv2 restorer instead of DKP | x2 super-resolution · DIV2KRK | Y-PSNR: 23.18 -> 31.86 | Improved |
| plug learned kernel into DANv2 restorer instead of ZSSR (ablation) | x2 super-resolution · DIV2KRK | Y-PSNR: 31.61 -> 31.86 | Improved |
| use estimated kernel instead of KernelGAN | x2 super-resolution · DIV2KRK | kernel PSNR: 51.5 -> 64.2 | Improved |
| run second stage with 5 random seeds | Ours + DiffPIR · FFHQ motion blur | PSNR: 0.02 | No clear effect |
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.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · 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
- Fast samplers for Inverse Problems in Iterative Refinement models (NeurIPS 2024 · similarity 0.89)
- Exact Posterior Score Estimation for Solving Linear Inverse Problems (NeurIPS 2026 · similarity 0.88)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.88)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.88)
- A Diffusion Model with State Estimation for Degradation-Blind Inverse Imaging (AAAI 2024 · similarity 0.87)
- Improving Diffusion Posterior Samplers with Lagged Temporal Corrections for Image Restoration (NeurIPS 2026 · similarity 0.87)
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.
