IncepHoloRGB: multi-wavelength network model for full-color 3D computer-generated holography
Xuan Yu, Zhilin Teng, Xuhao Fan, Tianchi Liu, Wenbin Chen, Xing-er Wang, Zhe Zhao, Wei Xiong, Hui Gao
Abstract
The popularity of deep learning has boosted computer-generated holography (CGH) as a vibrant research field, particularly physics-driven unsupervised learning. Nevertheless, present unsupervised CGH models have not yet explored the potential of generating full-color 3D holograms through a unified framework. In this study, we propose a lightweight multi-wavelength network model capable of high-fidelity and efficient full-color hologram generation in both 2D and 3D display, called IncepHoloRGB. The high-speed simultaneous generation of RGB holograms at 191 frames per second (FPS) are based on Inception sampling blocks and multi-wavelength propagation module integrated with depth-traced superimposition, achieving an average structural similarity (SSIM) of 0.88 and peak signal-to-noise ratio (PSNR) of 29.00 on the DIV2K test set in reconstruction. Full-color reconstruction of numerical simulations and optical experiments show that IncepHoloRGB is versatile to diverse scenarios and can obtain authentic full-color holographic 3D display within a unified network model, paving the way for applications towards real-time dynamic naked-eye 3D display, virtual and augmented reality (VR/AR) systems.