AI RESEARCH
Coloring the Noise: Adversarial Sobolev Alignment for Faithful Image Super Resolution
arXiv CS.AI
•
ArXi:2605.23264v1 Announce Type: cross Generative priors in Image Super-Resolution (SR) often compromise faithful restoration, we attribute this limitation to a fundamental spectral misalignment between isotropic objectives and the intrinsic natural image manifold. While Direct Preference Optimization offers a path to alignment, its reliance on spectrally flat Gaussian noise fails to distinguish authentic high-frequency details from hallucinations.