Running summarizing tools on a new article

Abstract: In this work, we propose a machine learning-based pipeline for generating a 3D-colored mesh representation of a scene from a single-input panorama. Our approach allows us to create a fully immersive Virtual Reality (VR) experiences. Unlike neural rendering approaches that are constrained by lowresolution inputs and offline processing, we focus on high-quality resolution inputs (4 K) and real-time rendering on VR headsets by estimating a depth map and creating a layered depth image (LDI) representation. We iteratively fill unseen regions behind objects through a cutting and inpainting process, after which the LDI is converted into an optimized, texture-mapped 3D mesh, resulting in a compact and efficient representation. We assess the feasibility of this approach through a quantitative evaluation with a user study in VR, focusing on the practicality, visual quality, and perceived presence of the generated scenes.