Reproduce “Anchored Inertial Proposals with Objective Verification for Diffusion-Prior Inverse Problems” 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
Minwoo Kim, Seunghyeok Shin, Dabin Kim, Hongki Lim
PIVOT reconstructs images from measurements using a pretrained diffusion model, proposing corrections and checking each against a frozen stepwise objective. Heavy-ball proposals improve PSNR over plain-gradient proposals in two tested ablations. On FFHQ random inpainting, it has mean per-image PSNR of 32.641 dB versus 31.022 for FAST-DIPS, but runs slower; verification guarantees nonincrease only for the current objective.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.