Research map / Image generation / Image restoration
Zero-Shot Adaptation for Approximate Posterior Sampling of Diffusion Models in Inverse Problems
Yasar U Alcalar, Mehmet Akcakaya
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 irregular noise schedule instead of uniform schedule (ablation) | ZAPS · FFHQ super-resolution x4, NFE=300 | LPIPS: 0.119 -> 0.104 | Improved |
| use irregular noise schedule instead of uniform schedule (ablation) | ZAPS · FFHQ random inpainting 70%, NFE=500 | PSNR: 26.94 -> 28.11 | Improved |
| use irregular vs uniform schedule at small NFE budget (ablation) | ZAPS · FFHQ super-resolution x4, NFE=100 | PSNR: 25.92 -> 26.08 | No clear effect |
| add zero-shot learned log-likelihood weights and diagonal Hessian approximation to DPS | DPS · FFHQ super-resolution x4 | PSNR: 23.86 -> 26.63 | Improved |
| replace DPS 1000 steps with ZAPS 300 NFEs | DPS · wall-clock time | WCT (s): 47.25 -> 14.71 | Improved |
| vary total timesteps S (20, 30, 60) at fixed NFE of 300 (ablation) | ZAPS · FFHQ super-resolution x4 | reconstruction quality | No clear effect |
| use DDPM instead of DDIM sampling within ZAPS (ablation) | ZAPS · FFHQ reconstruction | visual and quantitative quality | Improved |
| ZAPS versus PiGDM on random inpainting SSIM | ZAPS · FFHQ random inpainting 70% | SSIM: 0.842 -> 0.813 | Hurt |
| ZAPS versus DDRM on ImageNet super-resolution | ZAPS · ImageNet super-resolution x4 | PSNR: 24.33 -> 23.82 | Hurt |
| ZAPS versus DDNM and DiffPIR on LPIPS | ZAPS · FFHQ Gaussian deblurring, NFE=100 | LPIPS: 0.172 -> 0.128 | Improved |
Closest papers by meaning
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.93)
- InvFusion: Bridging Supervised and Zero-shot Diffusion for Inverse Problems (NeurIPS 2025 · similarity 0.91)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.90)
- Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors (NeurIPS 2024 · similarity 0.90)
- Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing (CVPR 2025 · similarity 0.90)
- Improving Diffusion Models for Inverse Problems Using Optimal Posterior Covariance (ICML 2024 · similarity 0.90)
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.
