Research map / Image generation / Fast diffusion sampling
Solving Diffusion Inverse Problems with Restart Posterior Sampling
Bilal Ahmed, Joseph Makin
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 restarts to the conditioned ODE (RePS) versus plain ODE or SDE (ablation) | RePS · FFHQ Gaussian deblur | PSNR and LPIPS across NFEs | Improved |
| use plain conditioned ODE without restart at high NFE (ablation) | conditioned ODE · FFHQ Gaussian deblur | PSNR and LPIPS | Hurt |
| increase conditioned ODE steps between restarts up to about 5 steps (ablation) | RePS · inverse problems, 1000 NFEs | PSNR | Improved |
| increase ODE steps between restarts beyond about 20 (ablation) | RePS · inverse problems, 1000 NFEs | LPIPS | Hurt |
| use multiple conditioned ODE steps per restart instead of one (DiffPIR-style) (ablation) | RePS · linear inverse problems | PSNR/LPIPS | Improved |
| replace DAPS two-stage conditioning with conditioned ODE (RePS) | RePS vs DAPS · linear and nonlinear tasks | PSNR/LPIPS | Mixed |
| RePS vs DAPS on motion deblurring across NFEs | RePS · motion deblurring | PSNR and LPIPS | Improved |
| RePS vs DAPS sampling time | RePS · all tasks | time per image | No clear effect |
Closest papers by meaning
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.93)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.92)
- Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing (CVPR 2025 · similarity 0.92)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.92)
- SITCOM: Step-wise Triple-Consistent Diffusion Sampling For Inverse Problems (ICML 2025 · similarity 0.92)
- Coupled Data and Measurement Space Dynamics for Enhanced Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.92)
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.
