Computer-Generated Hologram Based on 3d Gaussian Splatting

Abstract

In this study, we proposed a 3D hologram generation pipeline that use 2D images. In contract to approaches that require specialized depth cameras, our method captures 2D images with ordinary cameras or computer graphics rendering and then generates photorealistic 3D scenes using 3D Gaussian splatting. The reconstructed 3D scenes are then directly input into tensor holography to generate high-resolution holograms rapidly. Tensor holography leverages neural networks to produce holograms efficiently, thereby avoiding the need for computationally intensive diffraction calculations. Thus, the proposed pipeline integrates neural networks and optimization processes, eliminating the need for time-consuming processes such as physical diffraction computations.

Keywords: Computer-Generated Holography, 3D Gaussian Splatting, Tensor Holography, Neural Networks