Reproduce “Score-Based Diffusion Models as Principled Priors for Inverse Imaging” 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
Berthy T. Feng, Jamie Smith, Michael Rubinstein, Huiwen Chang, Katherine L. Bouman, William T. Freeman
The authors turn a score-based diffusion model into an image prior for Bayesian inverse imaging. A probability-flow ODE computes image log-probability, which Deep Probabilistic Imaging combines with measurement likelihoods through a RealNVP flow to approximate the posterior. On CelebA deblurring, averaged over 128 samples, PSNR was 24.75 versus DPS’s 20.37 using the lowest 6.25% of DFT frequencies and noise |σ|=1.0. The method remains approximate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.