Propagation as Data (PaD): Neural Phase Hologram Generation with Variable Distance Support

Jun Yeong Cha, Hyunmin Ban, Seongmi Choi, Hui Yong Kim

Abstract: Most of the neural network models for generating phase holograms developed so far are trained and validated only for a single distance. Consequently, if a distance is altered, the performance of models tends to decline dramatically. To address this, we introduce a novel approach called ‘Propagation as Data (PaD)’. Unlike conventional methods, our proposed model does not include the propagation process in a neural network. We pre-calculate propagation kernels and use them as conditioning data. Experimental results demonstrate that our model can consistently generate high-quality phase holograms across a range of distances with a single model.