Table 4 Summary of all strategies in “DL-in-processing for phase recovery”
Strategy | Network task | Input | Output | Dataset | Learning mode |
|---|---|---|---|---|---|
Network-only by dataset-driven | Phase recovery | Hologram | Phase | Paired dataset | Supervised |
Unpaired dataset | Weak-supervised | ||||
Network-only by untrained physics-driven | Phase recovery | Hologram | Phase | No requirement | Unsupervised |
Network-only by trained physics-driven | Phase recovery | Hologram | Phase | Input-only dataset | Unsupervised |
Physics-connect-network | Artifacts removal | Initial phase | Phase | Paired dataset | Supervised |
Network-in-physics with denoisers | Regularization | Noisy phase | Phase | Paired dataset | Supervised |
Network-in-physics with structural priors | Regularization | Fixed vector | Phase | No requirement | Unsupervised |
Network-in-physics with generative priors | Regularization | Latent vector | Phase | Phase-only dataset | Unsupervised |
Physics-in-network with interpretability | Phase recovery | Hologram | Phase | Fewer paired dataset | Supervised |
