Research map / Image generation / Fast diffusion sampling
Improving Diffusion Models for Inverse Problems using Manifold Constraints
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 MCG gradient step to projection-based solver (ablation) | score-SDE projection solver · FFHQ box inpainting | LPIPS: 0.138 -> 0.089 | Helped |
| add projection step to MCG (ablation) | MCG · FFHQ box inpainting | LPIPS: 0.124 -> 0.089 | Helped |
| add projection step to MCG (ablation) | MCG · FFHQ box inpainting | measurement consistency MSE: 10.7 -> 0 | Helped |
| compute gradient on Tweedie-denoised x0 instead of noisy sample (ablation) | MCG · FFHQ box inpainting | LPIPS: 0.128 -> 0.089 | Helped |
| MCG vs Score-CT | diffusion CT reconstruction · AAPM 18 views | PSNR: 29.85 -> 33.57 | Helped |
| MCG vs Score-SDE | diffusion colorization · FFHQ colorization | LPIPS: 0.180 -> 0.146 | Helped |
| add MCG gradient step (backprop) | score-SDE · inpainting runtime | wall-clock time: 38.68 -> 81.59 | Hurt |
| vary MCG step size alpha (ablation) | MCG · FFHQ box inpainting | LPIPS | Mixed |
| reduce NFE below 100 (ablation) | MCG · FFHQ box inpainting | LPIPS | Hurt |
Closest papers by meaning
- SITCOM: Step-wise Triple-Consistent Diffusion Sampling For Inverse Problems (ICML 2025 · similarity 0.90)
- Diffusion State-Guided Projected Gradient for Inverse Problems (ICLR 2025 · similarity 0.89)
- Solving Inverse Problems with Latent Diffusion Models via Hard Data Consistency (ICLR 2024 · similarity 0.88)
- Linearly Constrained Diffusion Implicit Models (NeurIPS 2025 · similarity 0.88)
- Decomposed Diffusion Sampler for Accelerating Large-Scale Inverse Problems (ICLR 2024 · similarity 0.87)
- DMPlug: A Plug-in Method for Solving Inverse Problems with Diffusion Models (NeurIPS 2024 · 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.