{ "id": "epqe5xg2qv", "slug": "neural-field", "title": "Neural field", "updated": "2026-10-15", "topic_path": [ { "id": "technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/topics/technology" }, { "id": "technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing" }, { "id": "technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai" }, { "id": "technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.ml_computation" }, { "id": "technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep" }, { "id": "technology.computing.ai.neural_deep.alpha-i-to-z", "label": "Titles I to Z", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep.alpha-i-to-z" } ], "geo": [ { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Non-geographic · 2001 to 2020: Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep", "path": [ { "id": "geo.nongeo", "label": "Non-geographic", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo" }, { "id": "geo.nongeo.t2001", "label": "Non-geographic · 2001 to 2020", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001" }, { "id": "geo.nongeo.t2001.technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology" }, { "id": "geo.nongeo.t2001.technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing" }, { "id": "geo.nongeo.t2001.technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai" }, { "id": "geo.nongeo.t2001.technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.ml_computation" }, { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep" } ] } ], "excerpt": "A neural field, also known as an implicit neural representation or coordinate-based network, is a field parameterized by a neural network, popularized by NeRF in 2020 for novel view synthesis.", "snippet": "A neural field, also known as an implicit neural representation or coordinate-based network, is a field parameterized by a neural network, popularized by NeRF in 2020 for novel view synthesis.", "node": "technology.computing.ai.neural_deep.alpha-i-to-z", "markdown": "# Neural field\n\nA neural field is a field, meaning a quantity defined over a continuous domain such as an image, a volume, or space-time, that is parameterized fully or in part by a neural network, most often a coordinate-based [multilayer perceptron](https://www.edgepedia.org/multilayer-perceptron) (MLP) that maps input coordinates to signal values.[1](https://doi.org/10.1111/cgf.14505) Training the network on a reconstruction loss turns its weights into a compact, differentiable representation of the signal, which can then be queried at arbitrary coordinates or rendered through a differentiable forward map. The same idea appears under the names implicit neural representation, neural implicit, and coordinate-based network.\n\n| Key fact | Value |\n|---|---|\n| Definition | A field parameterized fully or in part by a neural network[1](https://doi.org/10.1111/cgf.14505) |\n| Memory scaling | With network parameters, not spatio-temporal resolution[2](https://ar5iv.labs.arxiv.org/html/2111.11426) |\n| NeRF scene size | 5 MB of weights, about 3000x smaller than LLFF's >15 GB per-scene grid[3](https://doi.org/10.48550/arxiv.2003.08934) |\n| Original NeRF cost | 150-200 million network queries per rendered image, roughly 30 s per frame on an NVIDIA V100; at least 12 h training per scene[3](https://doi.org/10.48550/arxiv.2003.08934) |\n| Fastest training | Instant-NGP trains graphics primitives in seconds and renders 1920x1080 in tens of milliseconds[4](https://arxiv.org/pdf/2201.05989) |\n| Explicit alternative | Plenoxels optimize a sparse voxel grid in 11 min (bounded scenes) with no neural components[5](https://openaccess.thecvf.com/content/CVPR2022/papers/Fridovich-Keil_Plenoxels_Radiance_Fields_Without_Neural_Networks_CVPR_2022_paper.pdf) |\n| Post-2023 shift | 3D Gaussian splatting renders 1080p in real time at high quality, trading memory for speed[6](https://ar5iv.labs.arxiv.org/html/2308.04079) |\n\n## How it works\n\nA neural field is a function \\( f_{\\theta} \\) that takes coordinates \\( \\mathbf{v} \\) (for example \\( (x,y) \\) in an image or \\( (x,y,z) \\) plus viewing direction in a scene) and returns signal values such as color, density, or a signed distance. The parameters \\( \\theta \\) are fit by gradient descent on a reconstruction loss against observed data. Because the network is continuous and differentiable, memory scales with the number of parameters rather than with spatio-temporal resolution, unlike discrete grids whose size is limited by the Nyquist sampling rate of the signal.[2](https://ar5iv.labs.arxiv.org/html/2111.11426)\n\n**Spectral bias** is the central obstacle. Standard MLPs have difficulty learning high-frequency functions, a phenomenon the literature calls spectral bias.[7](https://arxiv.org/pdf/2006.10739) In the neural tangent kernel (NTK) view, introduced by Jacot, Gabriel, and Hongler in 2018,[8](https://bmild.github.io/fourfeat/index.html) training behaves like kernel regression: the \\( i \\)-th error component decays approximately exponentially at rate \\( \\eta \\lambda_{i} \\), where \\( \\lambda_{i} \\) is an NTK eigenvalue.[7](https://arxiv.org/pdf/2006.10739) A conventional MLP's NTK eigenvalues decay rapidly with frequency, so high-frequency content is learned last or not at all.\n\n**Fourier feature mappings** fix this by preprocessing coordinates. The mapping sends an input \\( \\mathbf{v} \\) to\n\n\\[ \\gamma(\\mathbf{v}) = \\left[ a_{1}\\cos(2\\pi \\mathbf{b}_{1}^{T}\\mathbf{v}),\\ a_{1}\\sin(2\\pi \\mathbf{b}_{1}^{T}\\mathbf{v}),\\ \\ldots,\\ a_{m}\\cos(2\\pi \\mathbf{b}_{m}^{T}\\mathbf{v}),\\ a_{m}\\sin(2\\pi \\mathbf{b}_{m}^{T}\\mathbf{v}) \\right]^{T} \\]\n\nbefore the MLP.[7](https://arxiv.org/pdf/2006.10739) With \\( a_{j}=1 \\) and the vectors \\( \\mathbf{b}_{j} \\) drawn from an isotropic Gaussian \\( \\mathcal{N}(0, \\sigma^{2}) \\), the sinusoidal mapping transforms a dot-product kernel into a stationary one, \\( k(\\gamma(\\mathbf{u}), \\gamma(\\mathbf{v})) = \\tilde{h}(\\mathbf{u}-\\mathbf{v}) \\), with a tunable bandwidth suited to low-dimensional domains.[7](https://arxiv.org/pdf/2006.10739) The scale of the frequency distribution matters far more than its shape.[7](https://arxiv.org/pdf/2006.10739) The positional encoding used by NeRF,\n\n\\[ \\gamma(p) = \\left( \\sin(2^{0}\\pi p), \\cos(2^{0}\\pi p), \\ldots, \\sin(2^{L-1}\\pi p), \\cos(2^{L-1}\\pi p) \\right), \\]\n\nis the special case with log-linearly spaced frequencies, and it was described as crucial for recovering high-frequency detail.[7](https://arxiv.org/pdf/2006.10739) • [9](https://cseweb.ucsd.edu/~ravir/nerficbs.pdf) The frequency choice is a trade-off: lower encoding frequencies produce blurry reconstruction, while higher frequencies introduce salt-and-pepper artifacts.[2](https://ar5iv.labs.arxiv.org/html/2111.11426)\n\n## How it is done\n\nThe typical workflow proceeds as follows.[2](https://ar5iv.labs.arxiv.org/html/2111.11426)\n\n1. **Choose architecture and encoding.** Select an MLP size and an input encoding (positional encoding, Fourier features, sine activations, or a hash grid). Capacity knobs include encoding scale, network width, and hash-table size; Instant-NGP's encoding is configured by just two values, the number of parameters \\( T \\) and the finest resolution \\( N_{\\mathrm{max}} \\).[4](https://arxiv.org/pdf/2201.05989)\n2. **Sample coordinates.** Draw coordinates from the domain; for view synthesis, sample points and viewing directions along camera rays.[10](https://huggingface.co/learn/computer-vision-course/unit8/nerf)\n3. **Apply a differentiable forward map.** Feed coordinates through the network and relate the outputs to the sensor domain. In NeRF-style rendering, the discretized volumetric rendering equation combines sampled colors \\( \\mathbf{c}_{i} \\), densities \\( \\sigma_{i} \\), and spacings \\( \\delta_{i} \\):\n\n\\[ \\hat{C}(\\mathbf{r}) = \\sum_{i=1}^{N} T_{i}\\left(1 - \\exp(-\\sigma_{i}\\delta_{i})\\right)\\mathbf{c}_{i}, \\qquad T_{i} = \\exp\\left(-\\sum_{j=1}^{i-1}\\sigma_{j}\\delta_{j}\\right). \\]\n\n[Volume rendering](https://www.edgepedia.org/volume-rendering) is naturally differentiable, so optimization requires only images with known camera poses, typically from COLMAP.[10](https://huggingface.co/learn/computer-vision-course/unit8/nerf) • [9](https://cseweb.ucsd.edu/~ravir/nerficbs.pdf)\n4. **Define the loss.** A pixel-wise reconstruction loss such as \\( L_{\\mathrm{recon}}(\\hat{C}, C^{*}) = \\lVert \\hat{C} - C^{*} \\rVert^{2} \\) is standard;[10](https://huggingface.co/learn/computer-vision-course/unit8/nerf) derivative-based penalties such as the Eikonal loss are added for signed-distance fields.[11](https://proceedings.neurips.cc/paper_files/paper/2020/file/53c04118df112c13a8c34b38343b9c10-Paper.pdf)\n5. **Optimize and query.** Fit \\( \\theta \\) by gradient descent, then query the field at arbitrary coordinates or render it through the forward map.\n\n## Origin\n\nCoordinate-based networks representing fields gained significant attention from 2019 onward.[2](https://ar5iv.labs.arxiv.org/html/2111.11426) The idea builds on earlier work representing 3D shapes with signed distance functions and occupancy functions, and on random Fourier features, which approximate stationary kernels via Bochner's theorem; the Fourier feature paper by Tancik and colleagues (NeurIPS 2020) connects that lineage to coordinate networks.[7](https://arxiv.org/pdf/2006.10739) NeRF, reported by Mildenhall and colleagues in 2020, popularized the formulation for view synthesis,[3](https://doi.org/10.48550/arxiv.2003.08934) and its intellectual roots trace to the plenoptic function, a seven-dimensional description of light rays.[12](https://arxiv.org/abs/2304.10050) SIREN, reported by Sitzmann and colleagues in 2020, replaced ReLU with periodic sine activations.[11](https://proceedings.neurips.cc/paper_files/paper/2020/file/53c04118df112c13a8c34b38343b9c10-Paper.pdf) The survey by Xie and colleagues (Computer Graphics Forum, 2022) consolidated the term \"neural field\" and reviewed over 250 papers.[1](https://doi.org/10.1111/cgf.14505) No published source identifies who first coined the term \"neural field\"; the survey is the work that established it.\n\n## Variants\n\n**Activation and encoding variants.** SIREN uses sine activations, which converge fast enough to fit a single image in a few seconds on a modern GPU.[11](https://proceedings.neurips.cc/paper_files/paper/2020/file/53c04118df112c13a8c34b38343b9c10-Paper.pdf) Instant-NGP, reported by Müller, Evans, Schied, and Keller (ACM Transactions on Graphics, 2022), replaced frequency encodings with a trainable multiresolution hash encoding: at a similar trainable parameter count it trains over 8x faster than a frequency-encoding configuration, matches a dense grid's quality with 20x fewer parameters, and reaches PSNR competitive with NeRF and NSVF after 15 s of training.[4](https://arxiv.org/pdf/2201.05989) Mip-NeRF addresses NeRF's aliasing by tracing cones instead of rays with integrated positional encoding, replacing the coarse and fine MLPs with one multiscale MLP.[13](https://arxiv.org/pdf/2306.03000v3.pdf)\n\n**Factorized and explicit-grid variants.** Plenoxels store density and spherical harmonic coefficients per voxel in a sparse grid, with no neural components, and optimize bounded scenes in 11 minutes on a Titan RTX, a more than 100x speedup over NeRF's roughly one day; their authors conclude that the key element of NeRF is the differentiable volumetric renderer, not the neural network.[5](https://openaccess.thecvf.com/content/CVPR2022/papers/Fridovich-Keil_Plenoxels_Radiance_Fields_Without_Neural_Networks_CVPR_2022_paper.pdf) TensoRF, reported by Chen and colleagues (2022), factorizes the feature grid as a tensor, reducing space complexity from \\( O(n^{3}) \\) to \\( O(n) \\) with CP or \\( O(n^{2}) \\) with vector-matrix decomposition.[14](https://www.cvlibs.net/publications/Chen2022ECCV.pdf) K-Planes represents a \\( d \\)-dimensional scene with \\( d \\)-choose-2 planes, achieving 1000x compression over a full 4D grid.[15](https://openaccess.thecvf.com/content/CVPR2023/papers/Fridovich-Keil_K-Planes_Explicit_Radiance_Fields_in_Space_Time_and_Appearance_CVPR_2023_paper.pdf)\n\n## Applications\n\n**Novel view synthesis** is the flagship use: NeRF represents a scene as a continuous 5D function mapping location and viewing direction to density and view-dependent radiance, optimized from posed images.[3](https://doi.org/10.48550/arxiv.2003.08934) NeRF ideas have since spread to image reconstruction, super resolution, pose estimation, depth estimation, 3D-aware image synthesis, and neural rendering.[12](https://arxiv.org/abs/2304.10050)\n\n**Geometry** is represented through signed distance or occupancy functions; because any derivative of a SIREN is itself a composition of SIRENs, derivatives can be supervised directly for boundary value problems including the Eikonal, Poisson, Helmholtz, and wave equations.[11](https://proceedings.neurips.cc/paper_files/paper/2020/file/53c04118df112c13a8c34b38343b9c10-Paper.pdf)\n\n**Compression.** Implicit neural representations overfit a small network to a single signal, storing it in the weights. For video, NeRV was the first INR targeting video specifically, generating whole frames from \\( t \\) via convolution and upsampling, so that compression becomes model compression through training, pruning, weight quantization, and entropy coding.[16](https://link.springer.com/article/10.1007/s11042-026-21822-5) For medical data, an end-to-end SIREN-based architecture compresses multi-parametric MRI at up to 97.5%.[17](https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0314944)\n\n## Limitations and alternatives\n\nThe original formulation is slow: rendering one image requires 150 to 200 million network queries, about 30 seconds per frame on a V100, and single-scene training takes at least 12 hours.[3](https://doi.org/10.48550/arxiv.2003.08934) NeRF-based methods remain computationally intensive for high-resolution outputs, and editing is difficult because changes to network weights do not map intuitively to geometric or appearance changes.[18](https://arxiv.org/pdf/2401.03890.pdf) Implicit dynamic-scene models can be far costlier still: DyNeRF used 8 GPUs for one week to train a single scene.[15](https://openaccess.thecvf.com/content/CVPR2023/papers/Fridovich-Keil_K-Planes_Explicit_Radiance_Fields_in_Space_Time_and_Appearance_CVPR_2023_paper.pdf) Spectral bias persists as the most common reconstruction-quality issue in INRs, motivating patch-wise decoding, high-frequency additions, wavelet modules, and frequency-domain losses in video variants.[16](https://link.springer.com/article/10.1007/s11042-026-21822-5)\n\n**Explicit and hybrid alternatives** trade these weaknesses differently. 3D [Gaussian splatting](https://www.edgepedia.org/gaussian-splatting), reported by Kerbl, Kopanas, Leimkuehler, and Drettakis (ACM Transactions on Graphics, 2023), optimizes 1 to 5 million explicit Gaussians and rasterizes them with tile-based splatting for real-time 1080p rendering.[6](https://ar5iv.labs.arxiv.org/html/2308.04079) NeRF's ray-marching requires expensive stochastic sampling that can produce noise, whereas rasterization parallelizes well; the cost is memory.[6](https://ar5iv.labs.arxiv.org/html/2308.04079) Controlled comparisons favor each side in different regimes: Gaussian splatting performs well with plentiful, similar training views, while NeRFs work better when test views differ from training views, are more stable on in-the-wild data, recover better geometry with limited views, and are more compact.[19](https://arxiv.org/html/2405.09717v3)\n\nSince late 2023 the field has moved toward hybrid neural-explicit models. HyRF decomposes scenes into grid-based neural fields plus sparse explicit Gaussians holding only 8 parameters each.[20](https://proceedings.neurips.cc/paper_files/paper/2025/file/1a85a40d16f5142bd056eab5da6debdd-Paper-Conference.pdf) Published comparisons do not settle per-query latency of a plain coordinate MLP or the standardization status of INR codecs.\n\n## References\n\n1. [Yiheng Xie and colleagues (2022). Neural Fields in Visual Computing and Beyond. Computer Graphics Forum.](https://doi.org/10.1111/cgf.14505)\n2. [Neural Fields in Visual Computing and Beyond (Xie et al.; Eurographics STAR 2021 / Computer Graphics Forum 41(2):641-676, doi 10.1111/cgf.14505)](https://ar5iv.labs.arxiv.org/html/2111.11426)\n3. [Mildenhall, Ben and colleagues (2020). NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2003.08934)\n4. [Instant Neural Graphics Primitives with a Multiresolution Hash Encoding (Instant-NGP, SIGGRAPH 2022 / ACM TOG)](https://arxiv.org/pdf/2201.05989)\n5. [Plenoxels: Radiance Fields Without Neural Networks (CVPR 2022)](https://openaccess.thecvf.com/content/CVPR2022/papers/Fridovich-Keil_Plenoxels_Radiance_Fields_Without_Neural_Networks_CVPR_2022_paper.pdf)\n6. [3D Gaussian Splatting for Real-Time Radiance Field Rendering (Kerbl et al., SIGGRAPH 2023)](https://ar5iv.labs.arxiv.org/html/2308.04079)\n7. [Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains (Tancik et al., NeurIPS 2020)](https://arxiv.org/pdf/2006.10739)\n8. [Fourier Feature Networks (authors' project page)](https://bmild.github.io/fourfeat/index.html)\n9. [Introduction and Basic NeRF Algorithm (Ramamoorthi, CBS survey)](https://cseweb.ucsd.edu/~ravir/nerficbs.pdf)\n10. [Neural Radiance Fields (NeRFs) · Hugging Face tutorial](https://huggingface.co/learn/computer-vision-course/unit8/nerf)\n11. [Implicit Neural Representations with Periodic Activation Functions (SIREN), NeurIPS 2020](https://proceedings.neurips.cc/paper_files/paper/2020/file/53c04118df112c13a8c34b38343b9c10-Paper.pdf)\n12. [Neural Radiance Fields: Past, Present, and Future](https://arxiv.org/abs/2304.10050)\n13. [BeyondPixels: A Comprehensive Review of the Evolution of Neural Radiance Fields](https://arxiv.org/pdf/2306.03000v3.pdf)\n14. [TensoRF: Tensorial Radiance Fields (ECCV 2022)](https://www.cvlibs.net/publications/Chen2022ECCV.pdf)\n15. [K-Planes: Explicit Radiance Fields in Space, Time, and Appearance (CVPR 2023)](https://openaccess.thecvf.com/content/CVPR2023/papers/Fridovich-Keil_K-Planes_Explicit_Radiance_Fields_in_Space_Time_and_Appearance_CVPR_2023_paper.pdf)\n16. [A survey of implicit neural representations for video compression (Multimedia Tools and Applications, Springer)](https://link.springer.com/article/10.1007/s11042-026-21822-5)\n17. [An end-to-end implicit neural representation architecture for medical volume data (PLOS One)](https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0314944)\n18. [A Survey on 3D Gaussian Splatting (Chen and Wang)](https://arxiv.org/pdf/2401.03890.pdf)\n19. [From NeRFs to Gaussian Splats, and Back (He et al., UPenn GRASP Lab)](https://arxiv.org/html/2405.09717v3)\n20. [HyRF: Hybrid Radiance Fields for Memory-efficient and High-quality Novel View Synthesis (NeurIPS 2025)](https://proceedings.neurips.cc/paper_files/paper/2025/file/1a85a40d16f5142bd056eab5da6debdd-Paper-Conference.pdf)\n\n---\n*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Artificial intelligence and data › Machine learning and neural computation › Neural networks and deep learning › Titles I to Z*\n\n*Initially written Sep 29, 2026 · Edited: Oct 15, 2026*\n\n*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*\n\nLicense: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license\n", "same_as": [ "https://cseweb.ucsd.edu/~ravir/nerficbs.pdf" ], "url": "https://www.edgepedia.org/neural-field", "markdown_url": "https://www.edgepedia.org/neural-field.md", "license": { "name": "Edgepedia Community License 1.0", "url": "https://www.edgepedia.org/edgepedia/license", "summary": "Free with credit, commercial use included. AI training is open to everyone. For other uses, organizations over USD 100M in revenue or 100M monthly users license separately.", "spdx": "LicenseRef-Edgepedia-Community-1.0" }, "credit": "\"Neural field\", Edgepedia (EdgeChat), https://www.edgepedia.org/neural-field. Edgepedia Community License 1.0.", "credit_md": "\"[Neural field](https://www.edgepedia.org/neural-field)\", Edgepedia (EdgeChat), [https://www.edgepedia.org/neural-field](https://www.edgepedia.org/neural-field). [Edgepedia Community License 1.0](https://www.edgepedia.org/edgepedia/license).", "credit_html": "\"Neural field\", Edgepedia (EdgeChat), https://www.edgepedia.org/neural-field. Edgepedia Community License 1.0.", "speakable": "A neural field, also known as an implicit neural representation or coordinate-based network, is a field parameterized by a neural network, popularized by NeRF in 2020 for novel view synthesis." }