Multi-degradation-adaptation network for fundus image enhancement with degradation representation learning

Ruoyu Guo, Yiwen Xu, Anthony Tompkins, Maurice Pagnucco, Yang Song

Abstract: Highlights•A model for effectively restoring multiple unknown degradations in fundus images.•The model can dynamically generate filter weights for restoration during inference.•The model proposes degradation representation learning to improve generalisability.•The model can effectively restore both synthetic and real-world fundus images.