Advancements in Radiance Field Techniques for Volumetric Video Generation: A Technical Overview

Video consumption has grown rapidly worldwide, supported by advanced devices such as smartphones, tablets, and immersive headsets like Apple Vision Pro, which enable real-time 6 Degrees of Freedom (6DoF) navigation. Yet, the lack of engaging volumetric content limits the applications of these technologies in training and entertainment. Volumetric video offers promise but faces challenges in natural 3D+t reconstruction, coding, and rendering due to high computational demands. In 2020, the groundbreaking Neural Radiance Field (NeRF) paper introduced a new method for generating natural, free-viewpoint renderings of real scenes. Follow-up research, including 3D Gaussian Splatting, has improved efficiency and flexibility; however, most approaches still require separate models per frame, which complicates video representation. Recent extensions utilize temporal redundancy to create compact, consistent, and editable volumetric video. This paper reviews state-of-the-art radiance field-based methods, analyzing their strengths and limitations, and provides an objective evaluation using a diverse multi-view dataset, with a focus on applications for entertainment and training.