Unsupervised Deep Learning for Phase Retrieval via Teacher-Student Distillation

Phase retrieval (PR) is a challenging nonlinear inverse problem in scientific imaging that involves reconstructing the phase of a signal from its intensity measurements. Recently, there has been an increasing interest in deep learning-based PR. However, collecting ground-truth (GT) images are challenging in many domains, which motivates us to study an unsupervised learning approach for PR. This approach trains a end-to-end deep model without any GT image via a teacherstudent online distillation framework. Specifically, a teacher model is trained using a self-expressive loss with noise resistance, while a student model is trained with a consistency loss on augmented data to improve the teacher's dark knowledge. Additionally, we develop an enhanced unfolding network for both the teacher and student models. Extensive experiments show that our proposed approach outperforms existing unsupervised PR methods with higher computational efficiency and performs competitively against supervised methods.