NeRF: Neural Radiance Field in 3D Vision: A Comprehensive Review (Updated Post-Gaussian Splatting)

TL;DR

This survey chronicles the progression of Neural Radiance Fields (NeRF) from its 2020 inception through the Gaussian Splatting era and beyond (2023–2025). It contrasts implicit/hybrid neural field rendering with explicit point-based approaches, analyzes theory, datasets, benchmarks, and a broad spectrum of applications from urban reconstruction to human avatars. The work highlights key advances in view synthesis quality, training/inference speed, and data efficiency before GS, and documents post-GS developments, including diffusion-based generation, semantic grounding, and neural SLAM. It concludes that while Gaussian Splatting has shifted momentum toward faster, high-fidelity rendering, NeRF-like methods remain vital for implicit representations, volumetric rendering, and certain 3D understanding tasks, and it provides a detailed reference framework for current and future research.

Abstract

In March 2020, Neural Radiance Field (NeRF) revolutionized Computer Vision, allowing for implicit, neural network-based scene representation and novel view synthesis. NeRF models have found diverse applications in robotics, urban mapping, autonomous navigation, virtual reality/augmented reality, and more. In August 2023, Gaussian Splatting, a direct competitor to the NeRF-based framework, was proposed, gaining tremendous momentum and overtaking NeRF-based research in terms of interest as the dominant framework for novel view synthesis. We present a comprehensive survey of NeRF papers from the past five years (2020-2025). These include papers from the pre-Gaussian Splatting era, where NeRF dominated the field for novel view synthesis and 3D implicit and hybrid representation neural field learning. We also include works from the post-Gaussian Splatting era where NeRF and implicit/hybrid neural fields found more niche applications. Our survey is organized into architecture and application-based taxonomies in the pre-Gaussian Splatting era, as well as a categorization of active research areas for NeRF, neural field, and implicit/hybrid neural representation methods. We provide an introduction to the theory of NeRF and its training via differentiable volume rendering. We also present a benchmark comparison of the performance and speed of classical NeRF, implicit and hybrid neural representation, and neural field models, and an overview of key datasets.

Figures (18)

• Figure 1: Timeline of important and influential NeRF and neural volume rendering methods pre- and post-Gaussian Splatting. Above the line are general or fundamental methods used by future subsequent methods to build neural fields. Below the line are various application-specific methods. Post April 2022 and especially after Gaussian Splatting, there is a lack of influential fundamental neural field models.
• Figure 2: The NeRF volume rendering and training process (image sourced from nerf2020_mildenhall). (a) illustrates the selection of sampling points for individual pixels in a to-be-synthesized image. (b) illustrates the generation of densities and colors at the sampling points using NeRF MLP(s). (c) and (d) illustrate the generation of individual pixel color(s) using in-scene colors and densities along the associated camera ray(s) via volume rendering, and the comparison to ground truth pixel color(s), respectively.
• Figure 3: Original NeRF nerf2020_mildenhall results on four scenes of the NeRF Synthetic Dataset. Left to right: Ground truth, NeRF nerf2020_mildenhall, LLFF 2019llf_forwardfacingdataset, SRN 2019srn, NV 2019neuralvolumes.
• Figure 4: Diagram of the Integrated Positional Encoding (IPE) of mip-NeRF (Figure 1 in 2021mipnerf). a) Standard ray-based point sampled of NeRF; b) Cone-sampling of mip-NeRF using IPE, approximating conic frustums with multivariate Gaussian distributions.
• Figure 5: Taxonomy of selected key NeRF innovation papers. The papers are selected using a combination of citations and GitHub star rating. We note that the MLP-less speed-based models are not strictly speaking NeRF models. Nonetheless, we decided to include them in this taxonomy tree due to their recent popularity and their similarity to speed based NeRF models.