NeRF: Representing scenes as neural radiance ﬁelds for view synthesis

By Ben Mildenhall, Google Research, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi

3,943 citations recorded by the source corpus.

Subfields: Computer Graphics and Computer-Aided Design, 3D Vision & Pose

View Paper: https://www.semanticscholar.org/paper/cc08e072ce806ecc84d8bb6e65f896134306d8d1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Graphics and Computer-Aided Design

In 3D Vision & Pose

Citations in the corpus

One-2-3-45: Any Single Image to 3D Mesh in 45 Seconds without Per-Shape Optimization (2023, neutral): “Traditional NeRF-like methods [48, 74] use a neural network to represent a single scene and require per-scene optimization.”
Perceiver: General Perception with Iterative Attention (2021, neutral): “…structures in our architecture by associating position and modality-speciﬁc features with every input element (e.g. every pixel, or each audio sample) – these can be learned or constructed using high-ﬁdelity Fourier features (Mildenhall et al., 2020; Tancik et al., 2020; Vaswani et al., 2017).”
SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes (2023, neutral): “To derive a compact motion representation, we introduce a set of sparse control points Here, p i denotes the learnable coordinate of control point i in the canonical space. o i is the learnable radius parameter of a radial-basis-function (RBF) kernel that controls how the impact of a control point…”
Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields (2023, neutral): “In Neural Radiance Fields (NeRF), a neural network is trained to model a volumetric representation of a 3D scene such that novel views of that scene can be rendered via raytracing [21].”
Continuous 3D Perception Model with Persistent State (2025, neutral): “…traditional methods such as Structure from Motion (SfM) [1, 2, 28, 47, 65, 66, 70, 71, 80] and Simultaneous Localization and Mapping (SLAM) [11, 16, 19, 22, 37, 51, 53], as well as more recent approaches such as Neural Radiance Fields (NeRF) [12, 24, 49, 50, 84] and 3D Gaussian Splatting [36].”
Shap-E: Generating Conditional 3D Implicit Functions (2023, positive): “While Mildenhall et al. [38] use two separate NeRF models for the coarse and ﬁne rendering stages, we instead share the parameters between the two stages but use separate output heads for the coarse and ﬁne densities and colors.”
A Survey on 3D Gaussian Splatting (2024, neutral): “While NeRF excelled in creating photorealistic images, the need for faster, more efficient rendering methods was becoming increasingly apparent, especially for applications ( e.g ., virtual reality and autonomous driving) that are highly sensitive to latency.”
Baking Neural Radiance Fields for Real-Time View Synthesis (2021, positive): “NeRF [29] We directly use the results reported in the original paper by Mildenhall et al .”
iMAP: Implicit Mapping and Positioning in Real-Time (2021, positive): “Following the network architecture in NeRF [15], we use an MLP with 4 hidden layers of feature size 256 , and two output heads that map a 3D coordinate p = ( x, y, z ) to a colour and volume density value: F θ ( p ) = ( c , ρ ) .”
Extracting Triangular 3D Models, Materials, and Lighting From Images (2021, positive): “In addition, we adopt the positional encoding in NeRF [12] and progressively fit the frequencies similar to SAPE [6].”
LERF: Language Embedded Radiance Fields (2023, positive): “Neural Radiance Fields (NeRFs) [24] have emerged as a powerful technique for capturing photorealistic digital representations of intricate real-world 3D scenes.”
SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM (2023, positive): “The shortcomings of handcrafted representations, combined with the emergence of radiance field representations [21] for high-quality image synthesis, have fueled a recent class of methods that attempt to encode the scene into the weight space of a neural network.”
Efficient Geometry-aware 3D Generative Adversarial Networks (2021, positive): “These quantities are rendered into RGB images using (neural) volume rendering [41,45].”
DreamLLM: Synergistic Multimodal Comprehension and Creation (2023, positive): “Motivated by the great success in 2D, a series of works have significantly propelled the 3D synthesis development (Mildenhall et al., 2022; Liu et al., 2023d; Lin et al., 2023; Wang et al., 2023c; Tang et al., 2023) based on Score Distillation Sampling (SDS) (Poole et al., 2023; Wang et al., 2023a)…”
KITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3D (2021, positive): “For all NeRF-based methods [7], [27], [64], we train one model on each scene individually, using cascaded sampling with 256 coarse samples and 256 ﬁne samples.”