Fig. 1. Network framework and training principle of POH generation algorithms based on data-driven deep learning

Fig. 2. Residual U-Net architecture

Fig. 3. Different generation methods of POH datasets

Fig. 4. 3D scene inputs and corresponding optical reconstructions of TensorHolo v2 ^［24］

Fig. 5. Network framework and training principle of POH generation algorithms based on model-driven deep learning

Fig. 6. Network framework and training principle of the two-step model-driven deep learning method

Fig. 7. End-to-end network framework and optical reconstructions of Holo-Encoder^［35］

Fig. 8. Upsampling block and optical reconstructions of 4K-DMDNet^［36］

Fig. 9. Learning-based diffraction model called CNNprop-CNN^［42］