Reproduce “The Domain Is a Residue: Adapting Self-Supervised Features, Not Generators” 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
Thomas Deixelberger, Markus Steinberger
RFA translates scenes between domains by mapping frozen DINO feature maps before a frozen, feature-conditioned decoder renders images. The authors report that it removes wet appearance in rain examples while retaining recognizable layout, and reaches KID 0.0123 and FID 107.55 on PreSIL-to-Mapillary, versus 0.0331 and 125.35 for HyPER-GAN. However, RFA’s structure distance is higher, indicating scene changes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.