P‐3.4: Plenoxels‐based Parallax Map Generation for Flexible Scale Ultra‐High Resolution in 3D Imaging.

This study is devoted to the development of a neural network‐free radial field flexible scale ultra‐high resolution parallax map generation method to improve the generation of ultra‐high resolution parallax maps for naked eye 3D display. In naked eye 3D display, the acquisition of parallax maps is crucial, the traditional method is to shoot the object by camera array, which requires high hardware conditions, in recent years, with the advancement of technology, the generation of parallax maps by Nerf has appeared, but this method generates lower quality images and requires a longer time to train the model, with a single GPU training taking more than one hour. In this paper, we propose a method to generate parallax maps based on neural network‐free radial field and arbitrary scale super‐resolution, firstly, we generate sparse voxel mesh by neural network‐free radial field (Plenoxels), and the optimization time on a single Titan RTX GPU is more than ten minutes, and then we get parallax maps by setting up a virtual camera, however, the parallax maps obtained are low‐resolution and can't match with the size of the naked‐eye 3D displays, therefore, we introduce an arbitrary scale super‐resolution network for the parallax maps, and we get high‐resolution parallax maps and can match the size of the naked‐eye 3D displays. [ABSTRACT FROM AUTHOR]