Domain Adaptation for Phase Retrieval

Name

In this signature work, I design a new model that integrates data correlation, domain

label, and Fourier magnitude to perform unsupervised domain adaptation for phase

retrieval. To achieve robust image recovery from Fourier magnitude, we propose four

effective modules: encoder, decoder, domain alignment, and correlation alignment.

These modules can mutually reinforce and supplement each other to learn domain‐

invariant representations for phase retrieval. Experiments on the MNIST dataset are

carried out to validate the efficacy of our proposed model. The findings reveal that

our model surpasses current leading domain adaptation methods in phase retrieval

and offers valuable insights for upcoming research in this area.