Phase Retrieval Using Expectation Consistent Signal Recovery Algorithm Based on Hypernetwork

Phase retrieval (PR) is an important component in modern computational imaging systems. Many algorithms have been developed over the past half-century. Recent advances deep learning introduced new possibilities for a robust and fast PR. An emerging technique called unfolding provides systematic connection between conventional model-based iterative data-based learning. Unfolded algorithms, which are powered by data learning, shown remarkable performance convergence speed improvement original algorithms. Despite their potential, most existing unfolded strictly confined to fixed number of iterations when layer-dependent parameters used. In this study, we develop novel framework overcome limitations. Our development based on generalized expectation consistent signal recovery (GEC-SR) algorithm, wherein damping factors left data-driven particular, introduce hypernetwork generate GEC-SR. Instead set optimal directly, learns how according clinical settings, thereby ensuring its adaptivity different scenarios. To enable adapt varying layer numbers, use recurrent architecture dynamic that generates factor can vary online across layers. We also exploit self-attention mechanism enhance robustness hypernetwork. Extensive experiments show proposed algorithm outperforms ones terms accuracy still works well under very harsh even many classical PR unstable.