Solving Diffusion Inverse Problems with Restart Posterior Sampling
Bilal Ahmed, Joseph G. Makin · 2025
Read Solving Diffusion Inverse Problems with Restart Posterior Sampling by Bilal Ahmed, Joseph G. Makin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-20705.txt.
Description
Bilal Ahmed and Joseph G. Makin propose in 2025 a cheaper way to solve inverse problems with diffusion models, the image generators that start from noise and clean it up step by step. Inverse problems ask the reverse question: given partial or noisy measurements, recover the original signal. Diffusion models make good guides for this, but earlier methods either approximated crudely, needed costly gradient passes, or handled only linear measurements.
Their method, RePS, restarts the sampling process partway, shrinking the approximation errors that build up during generation. It never backpropagates through the score network, its noise predictor, removing a big cost. Tested on the FFHQ and ImageNet photo collections, RePS handled linear and nonlinear tasks. It enlarged images fourfold and filled a masked 128-by-128 box. It restored images with 70 percent of pixels hidden, and removed blur. Across these tasks it converged faster and reconstructed better than existing diffusion-based baselines.
Catalog record
| CREATOR | Bilal Ahmed, Joseph G. Makin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-20705