DVANet: Degradation-aware Visual-prior Alignment Network for Image Restoration
Researchers have developed DVANet, a novel deep unfolding network designed for unified image restoration across diverse degradation types. This network integrates a degradation-aware observation consistency module with a visual-prior-guided reconstruction branch, utilizing DINOv3 for structural and semantic information. DVANet aims to improve performance in suppressing degradation and recovering structural details, particularly in damaged regions, demonstrating strong adaptability and generalization capabilities in experiments. AI
IMPACT Introduces a new method for image restoration that could improve detail recovery and degradation suppression in various scenarios.