神经辐射场加速技术综述

摘要：神经辐射场（NeRF）技术可以从2D 图像中学习场景的3D 隐式模型，并合成出高清逼真新视角图像。该技术有着良好的应用前景，受到业界广泛关注。针对NeRF 技术运算缓慢的问题，近两年业界研究者提出了各种加速技术。对现有加速技术进行了综述，分类梳理并分析了速度提升背后的技术机理和工程技巧，同时讨论了未来加速技术演进的方向。本研究有助于激发更高效算法的产生，从而推进NeRF 技术在视觉内容生成及其他领域的应用。

关键词：NeRF；神经渲染；视点合成；体渲染

Abstract: The neural radiance field (NeRF) technology, which can learn the 3D implicit representation of a scene from a set of 2D images and synthesize high-resolution and photo-realistic images of novel views, has aroused extensive research interest due to its vast application potential. In order to solve NeRF’s problem of slow running speed, various acceleration technologies have been proposed in recent two years. We review current acceleration technologies by categorizing and analyzing their technical mechanism and engineering skills. We also discuss directions for further acceleration. Our work will contribute to inspiring the invention of more efficient algorithms and promote NeRF’s application in multiple fields including visual content generation and beyond.

Keywords: NeRF; neural rendering; view synthesis; volume rendering