Neural-Network-Enabled Large Scale Adaptive Glasses-Free Foveated 3D Displays

Glasses-free 3D displays are emerging as a next-generation technology that will redefine interactions with the digital world. One challenge to deliver a truly immersive viewing experience is limited spatial/angular resolution, caused by distributing pixels on the display panel across numerous viewing angles, thereby restricting the display quality of each individual perspective. Here we demonstrate a high-performance glasses-free 3D display with adaptive light field reconstruction through a machine-learning-based design process. This inverse design method, developed using a voxel-based neural network, optimizes a large-scale flat optics element for effective light-field modulation to realize portable 3D displays with arbitrary view distributions. This system enables higher display resolution by increasing the density of views only where users typically focus their attention while reducing density of views in less critical regions to optimize the display quality. With this method, we achieved a 100-mm flat-optics element with 1.5 × 10¹⁰ phase-modulating subpixels (optical degrees of freedom), far exceeding the pixel count of a 4K panel. We constructed a glasses-free 3D display with a remarkable angular resolution up to 0.67 views per degree by a simple integration with an off-the-shelf purchased liquid crystal display, achieving a two-fold increase in display resolution with smoother motion parallax than conventional systems.