Reproduce “Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Roi Benita, Michael Elad, Joseph Keshet
The paper designs guidance weights for zero-shot diffusion reconstruction, illustrated by an FFHQ-256 face with 70% of pixels missing. It derives weights from signal statistics, the degradation, and diffusion steps instead of hand-tuning them. In a low-pass filtering test, spectral DPS improves PSNR from 18.31 to 27.82 dB and FID from 62.68 to 41.31, but not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.