Reproduce “Getting Your Guidance Weights Right in diffusion and flow-matching posterior sampling” 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
Liam Moroy, Jean-François Giovannelli, Yoann Altmann, Steve McLaughlin, Frédéric Champagnat, Guillaume Bourmaud
SIMCA calibrates guidance weights for training-free image posterior sampling, balancing a pretrained network’s output against measurement guidance. At each timestep, it fits two scalar weights by least squares and propagates trajectories through states reached by its sampler. On FFHQ random inpainting, it improves PSNR, LPIPS, and FID over simulation-free calibration. At 50 steps, iterations fall from 1,000, while perceptual metrics worsen.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.