Efficient Diffusion Posterior Sampling for Noisy Inverse Problems

Ji Li, Chao Wang · 2025

Read Efficient Diffusion Posterior Sampling for Noisy Inverse Problems by Ji Li, Chao Wang (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-2503-10237.txt.

Description

Li and Wang speed up image recovery from noisy, partial data with diffusion models, in a paper from twenty twenty-five. Diffusion models trained on clean images can guide recovery with no retraining. The hard step is judging how likely a noisy draft is to match the real data. Older methods take gradients through the big model, which costs time and memory.

Their method swaps the gradient step for a best-fit estimate solved as a classic tuned problem. The problem balances fit to the data with fit to the learned prior. Plugged into a fast sampler, it solves both linear tasks and JPEG repair. Tests across many tasks show strong results. Both main classes of pre-trained diffusion models work with the method.

Catalog record

CREATORJi Li, Chao Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-10237

Browse the full library · About the Tower of Babel