Physical-aware uncertainty prompt learning for real-world blind image restoration

Yuanjian Qiao, Mingwen Shao, Lingzhuang Meng, Wangmeng Zuo

Abstract: Highlights•A novel PUP paradigm is proposed for real-world blind image restoration.•A Physical-aware Degradation Modeling is designed to generate training data online.•An Uncertainty-Prompted Fourier Transformer is developed for flexible restoration.•Achieved superior performance and model efficiency across various tasks.