Research map / Image generation / Image restoration
Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance
Jing Jia, Wei Yuan, Sifan Liu, Liyue Shen, Guanyang 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 |
|---|---|---|---|
| use a 3-step in-domain DDIM prior with initial-noise optimization instead of 1000-step DPS | initial-noise optimization · CelebA inpainting | PSNR: 31.98 -> 33.78 | Improved |
| use a bedroom-trained out-of-domain weak prior for CelebA inpainting vs DPS | initial-noise optimization · CelebA inpainting | PSNR: 31.98 -> 32.76 | Improved |
| use a 3-step prior with initial-noise optimization instead of DPS | initial-noise optimization · CelebA 4x super-resolution | PSNR: 26.82 -> 31.27 | Improved |
| use CelebA-trained prior instead of Church-trained prior (domain mismatch) | initial-noise optimization · Church inpainting | PSNR: 24.93 -> 23.56 | Hurt |
| use in-domain Bedroom prior instead of CelebA prior | initial-noise optimization · Bedroom inpainting | PSNR: 27.78 -> 28.88 | Improved |
| use our optimizer and early stopping instead of DMPlug | initial-noise optimization · CelebA inpainting | PSNR: 32.778 -> 33.784 | Improved |
| use our method instead of DMPlug on Church inpainting | initial-noise optimization · Church inpainting | PSNR: 24.957 -> 24.927 | No clear effect |
| add HoldoutTopK early stopping instead of variance-based stopping (ablation) | Adam · Bedroom inpainting | PSNR: 28.669 -> 29.156 | Improved |
| add HoldoutTopK early stopping instead of variance-based stopping (ablation) | AdamSphere · Bedroom inpainting | PSNR: 28.559 -> 29.162 | Improved |
| add AdamSphere in place of Adam (with HoldoutTopK) (ablation) | HoldoutTopK · Bedroom inpainting | PSNR: 29.156 -> 29.162 | No clear effect |
| add AdamSphere in place of Adam (with variance stopping) (ablation) | variance early stopping · Church Gaussian deblurring | PSNR: 22.034 -> 22.048 | No clear effect |
| increase box-inpainting masked fraction | weak prior vs DPS · Church box inpainting | LPIPS | Hurt |
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.
- DMPlug — DMPlug: A Plug-in Method for Solving Inverse Problems with Diffusion Models (NeurIPS 2024) · better on 3 of 3 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
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.89)
- Improving Diffusion Models for Inverse Problems Using Optimal Posterior Covariance (ICML 2024 · similarity 0.88)
- A Mixture-Based Framework for Guiding Diffusion Models (ICML 2025 · similarity 0.88)
- Learning Diffusion Priors from Observations by Expectation Maximization (NeurIPS 2024 · similarity 0.88)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.88)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (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.
