Reproduce “Lipschitz-Controlled Flow Matching for Robust Image Restoration” 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
Sedjro Salomon Hotegni, Konstantin Sonntag, Paul-Christian Bürkner, Sebastian Peitz
LipsFlower controls sensitivity in flow-matching models for blind face restoration. It combines a constrained velocity network with finite-radius gain regularization to limit amplification of input perturbations. On 3,000 CelebA-Test images, its Flow-from-Y variant reduced mean output gain from 0.52 to 0.47, CVaR95 from 0.849 to 0.747, and maximum gain from 1.74 to 1.16 under the tested RMS-budget SPSA attack.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.