49.3: A Modified Unsupervised Vision Transformer Network for High‐fidelity Computer‐generated Holography

Abstract

To generate high quality images with faster calculation speed for holographic displays, we propose a modified unsupervised Vision Transformer model, which has the capability of capturing globalfeatures of an image. The proposed method can infer a hologram significantly faster than the stochastic gradient descent method, while producing images with similar quality.