Reproduce “Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters” 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
Felix Nützel, Mischa Dombrowski, Luca Hagen, Bernhard Kainz
The authors introduce an offline adapter that reorganizes image latents for medical-image generation. It learns a reversible, distortion-controlled coordinate change while keeping the encoder and decoder fixed, aligning latent similarity structure with a teacher before generator training. On CT-RATE, CT-FID falls from 20.23 ± 0.16 to 15.46 ± 1.38, with balanced accuracy at 0.528. Evidence is limited to two settings and proxy metrics, not clinical validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.