Fig. 1. Schematic diagram of digital holographic recording

Fig. 2. Schematic of deep neural network

Fig. 3. Network architecture. (a) Convolutional neural network; (b) U-net^[38]

Fig. 4. Relationship between error between output and label in training set, test set, and validation set and number of iterations. (a) Training set and test set; (b) validation set

Fig. 5. Deep-learning-based hologram reconstruction. (a) Phase reconstruction of diffraction intensity map based on end-to-end deep learning^[53]; (b) removal of twin images in in-line holographic reconstruction based on deep learning^[28]; (c) deep-learning-based autofocusing and removal of twin-images with extended depth-of-field^[54]; (d) autofocus and extended depth of field in in-line holographic reconstruction based on deep learning^[31]; (e) amplitude and phase reconstruction of off-axis hologram based on deep learning^[55]

Fig. 6. Deep-learning-based hologram reconstruction with unpaired data^[57]. (a) Process of network training (X_i∈X: hologram; Y_i∈Y: phase image; G & F: generator; D_Y: determine the authenticity of the images generated by G; D_X: determine the authenticity of the images generated by F; L_cyc(D,F): loss of cyclic consistency; L_Y(G,D_Y,X,Y): antagonistic loss between D_Y and G; L_X(F,D_X,Y,X): antagonistic loss between D_X and F); (b) reconstruction results (I: untrained hologram of input network for testing; II, III: reconstruction based on traditional methods and deep learning; IV: real phase distribution)

Fig. 7. Deep-learning-based autofocusing and phase reconstruction. (a) Phase recovery of defocus hologram based on deep learning^[26]; (b) holographic self-focusing reconstruction based on end-to-end deep learning^[66]

Fig. 8. Deep-learning-based holographic reconstruction noise suppression. (a) Coherent noise suppression without clean data based on deep learning^[77]; (b) speckle noise suppression of in-line holographic reconstructed image based on deep learning^[25]; (c) off-axis hologram speckle noise suppression based on deep learning^[81]

Fig. 9. Resolution enhancement of hologram reconstruction based on deep learning^[82]

Fig. 10. Deep-learning-based noise-free quantitative phase imaging of in-line holography^[26]. (a) Method to achieve the noise-free quantitative phase image; (b) training dataset generation and network training process

Fig. 11. Deep-learning-based hologram generation^[89]

Fig. 12. Deep-learning-based cross-modality image transformations^[91]

Fig. 13. High-resolution numerical dark-field microscopy imaging based on deep learning^[98]