Research map / Image generation / Fast diffusion sampling
Solving Linear Inverse Problems Provably via Posterior Sampling with Latent Diffusion Models
Litu Rout, Negin Raoof, Giannis Daras, Constantine Caramanis, Alex Dimakis, Sanjay Shakkottai
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 |
|---|---|---|---|
| PSLD with Stable Diffusion instead of DPS on random inpainting | PSLD · FFHQ 256 random inpainting | FID: 33.48 -> 21.34 | Improved |
| PSLD instead of DPS on box inpainting | PSLD · FFHQ 256 box inpainting | FID: 35.14 -> 43.11 | Hurt |
| PSLD instead of DPS on box inpainting | PSLD · FFHQ 256 box inpainting | LPIPS: 0.216 -> 0.167 | Improved |
| PSLD instead of DPS on 4x super-resolution | PSLD · FFHQ 256 SR 4x | FID: 39.35 -> 34.28 | Improved |
| PSLD instead of DPS on Gaussian deblur | PSLD · FFHQ 256 Gaussian deblur | FID: 44.05 -> 41.53 | Improved |
| PSLD instead of DPS on super-resolution at 2x, 3x, 4x | PSLD · FFHQ 256 SR | LPIPS: 0.220 -> 0.185 (2x); 0.247 -> 0.220 (3x); 0.291 -> 0.233 (4x) | Improved |
| Use latent diffusion LDM-VQ-4 (same training data as DPS) for box inpainting | PSLD · FFHQ 256 box inpainting | PSNR: 23.39 -> 24.22 | Improved |
| Use LDM-VQ-4 for random inpainting versus DPS | PSLD · FFHQ 256 random inpainting | LPIPS: 0.212 -> 0.221 | No clear effect |
| Use LDM-VQ-4 for random inpainting versus DPS | PSLD · FFHQ 256 random inpainting | PSNR: 29.49 -> 30.31 | Improved |
| Use Stable Diffusion instead of LDM-VQ-4 as the foundation model | PSLD · FFHQ 256 random inpainting | LPIPS: 0.221 -> 0.096 | Improved |
| Random inpainting with added Gaussian noise sigma=0.05 (PSLD vs DPS) | PSLD · FFHQ 256 inpaint + denoise | PSNR: 29.49 -> 33.71 | Improved |
| Noiseless random inpainting sigma=0.00 (PSLD vs DPS) | PSLD · FFHQ 256 inpaint + denoise | LPIPS: 0.171 -> 0.083 | Improved |
Closest papers by meaning
- Solving Inverse Problems with Latent Diffusion Models via Hard Data Consistency (ICLR 2024 · similarity 0.92)
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.92)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.92)
- Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing (CVPR 2025 · similarity 0.92)
- Improving Diffusion Models for Inverse Problems Using Optimal Posterior Covariance (ICML 2024 · similarity 0.91)
- Prompt-tuning Latent Diffusion Models for Inverse Problems (ICML 2024 · similarity 0.91)
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.
