Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Radiance Fields for the Real World: A Survey

Wenhui Xiao, Remi Chierchia, Rodrigo Santa Cruz, Xuesong Li, David Ahmedt-Aristizabal, Olivier Salvado, Clinton Fookes, Leo Lebrat

TL;DR

<3-5 sentence high-level summary> NeRFs revolutionize 3D scene representation by learning continuous volumetric radiance fields from 2D images with differentiable volume rendering. The survey unifies fundamentals (sampling, encoding, rendering, and architectures) with real-world challenges (degraded views, pose errors, dynamic/indoor/unbounded scenes) and applications (reconstruction, robotics, recognition, 3D generation). It maps a broad landscape of techniques, from hierarchical sampling and hash-grid encodings to tensor decompositions and divide-and-conquer strategies, and discusses datasets, tools, and evaluation standards. The work also highlights open problems in uncertainty quantification, generalization to unseen scenes, and large-scale, real-time deployment, offering directions toward more robust, efficient, and multimodal neural rendering systems.

Abstract

Neural Radiance Fields (NeRFs) have remodeled 3D scene representation since release. NeRFs can effectively reconstruct complex 3D scenes from 2D images, advancing different fields and applications such as scene understanding, 3D content generation, and robotics. Despite significant research progress, a thorough review of recent innovations, applications, and challenges is lacking. This survey compiles key theoretical advancements and alternative representations and investigates emerging challenges. It further explores applications on reconstruction, highlights NeRFs' impact on computer vision and robotics, and reviews essential datasets and toolkits. By identifying gaps in the literature, this survey discusses open challenges and offers directions for future research.

Neural Radiance Fields for the Real World: A Survey

TL;DR

<3-5 sentence high-level summary> NeRFs revolutionize 3D scene representation by learning continuous volumetric radiance fields from 2D images with differentiable volume rendering. The survey unifies fundamentals (sampling, encoding, rendering, and architectures) with real-world challenges (degraded views, pose errors, dynamic/indoor/unbounded scenes) and applications (reconstruction, robotics, recognition, 3D generation). It maps a broad landscape of techniques, from hierarchical sampling and hash-grid encodings to tensor decompositions and divide-and-conquer strategies, and discusses datasets, tools, and evaluation standards. The work also highlights open problems in uncertainty quantification, generalization to unseen scenes, and large-scale, real-time deployment, offering directions toward more robust, efficient, and multimodal neural rendering systems.

Abstract

Neural Radiance Fields (NeRFs) have remodeled 3D scene representation since release. NeRFs can effectively reconstruct complex 3D scenes from 2D images, advancing different fields and applications such as scene understanding, 3D content generation, and robotics. Despite significant research progress, a thorough review of recent innovations, applications, and challenges is lacking. This survey compiles key theoretical advancements and alternative representations and investigates emerging challenges. It further explores applications on reconstruction, highlights NeRFs' impact on computer vision and robotics, and reviews essential datasets and toolkits. By identifying gaps in the literature, this survey discusses open challenges and offers directions for future research.
Paper Structure (54 sections, 10 equations, 3 figures, 3 tables)

This paper contains 54 sections, 10 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Our Contributions
  3. Scope and Organization
  4. Fundamentals of NeRF
  5. Sampling
  6. Encoding
  7. Radiance Fields Estimation
  8. Pre-caching
  9. Tensor rank decomposition/Factorization
  10. Divide-and-conquer
  11. Alternative Architectures
  12. Volume Rendering
  13. NeRF-Agnostic Enhancements
  14. Alternative Scene Representations
  15. Advancement Facing Real-world Challenges
  16. ...and 39 more sections

Figures (3)

  • Figure 1: Overview of our paper structure, outlining key sections covering NeRF fundamentals and improvement strategies, real-world challenges with corresponding solutions, diverse application domains, and practical resources. Images adapted from Barron2021MipNeRFdai2017scannetMildenhall2020NeRFblendernerfstudionerfactorjeong2021selfPark2021Nerfiesgoli2024bayescai2022pix2nerfneusTurki2022Mega-NeRFlerf2023poole2023dreamfusionkerbl23gaussiansplattingchen2021mvsnerfnerf-wd-nerfcunerfzhu2022nice.
  • Figure 2: Different multi-scale encoding designs.
  • Figure 3: