Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Local MAP Sampling for Diffusion Models
Shaorong Zhang, Rob Brekelmans, Greg Ver Steeg
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 LMAPS local MAP sampling instead of DAPS posterior sampling | DAPS · FFHQ 256 SR 4x | PSNR: 29.07 -> 30.74 | Improved |
| use LMAPS clean-space local MAP instead of DMAP latent-space local MAP | DMAP · FFHQ 256 Gaussian Deblurring | PSNR: 26.83 -> 30.88 | Improved |
| use LMAPS instead of DMAP | DMAP · FFHQ 256 SR 4x | FID: 44.78 -> 80.63 | Hurt |
| use LMAPS instead of DAPS on random inpainting | DAPS · FFHQ 256 Inpaint (Random) | PSNR: 31.12 -> 34.51 | Improved |
| use LMAPS instead of DAPS on motion deblurring | DAPS · FFHQ 256 Motion Deblurring | PSNR: 29.66 -> 32.62 | Improved |
| use LMAPS instead of RED-diff on nonlinear deblurring | RED-diff · FFHQ 256 Nonlinear Deblurring | PSNR: 30.86 -> 29.93 | Hurt |
| use LMAPS instead of SITCOM on ImageNet SR 4x | SITCOM · ImageNet 256 SR 4x | PSNR: 27.07 -> 26.72 | Hurt |
| use LMAPS instead of DDNM on ImageNet random inpainting | DDNM · ImageNet 256 Inpaint (Random) | PSNR: 31.16 -> 30.59 | Hurt |
| use LMAPS on scientific inverse problems versus best baseline | RED-diff / DAPS / DDNM · InverseBench LIS | PSNR: 36.56 -> 38.07 (NR=360) | Improved |
| use LMAPS instead of PnP-DM on CS-MRI 4x noiseless | PnP-DM · InverseBench CS-MRI | PSNR: 31.80 -> 32.83 | Improved |
| increase gradient optimization steps per diffusion step at fixed NFE budget (ablation) | LMAPS · Gaussian deblurring | reconstruction quality | Improved |
| use LMAPS in place of SITCOM, which backpropagates through the U-Net with 600 NFEs | SITCOM · Gaussian deblurring | performance vs compute | No clear effect |
Closest papers by meaning
- Rethinking Diffusion Posterior Sampling: From Conditional Score Estimator to Maximizing a Posterior (ICLR 2025 · similarity 0.90)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.90)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.90)
- Coupled Data and Measurement Space Dynamics for Enhanced Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.90)
- Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors (NeurIPS 2024 · similarity 0.89)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.89)
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.
