Deep learning for hologram generation

This work exploits deep learning to develop real-time hologram generation. We propose an original concept of introducing modulators allow the use generative models interpret complex-valued frequency data directly. new mechanism enables pre-trained model generate samples with variations in underlying features. To achieve object-based generation, we also a model, named channeled variational autoencoder (CVAE). The CVAE can then and learn hidden structure input holograms. It is thus able holograms through disentangled latent representations, which us specify each feature for specific object. Additionally, technique called super-resolution (HSR) super-resolve low-resolution output. Combining proposed HSR, successfully approach super-resolved, complex-amplitude 3D scenes.