Ben Mildenhall
Last reviewed
Jun 5, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 ยท 2,283 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 5, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 ยท 2,283 words
Add missing citations, update stale details, or suggest a clearer explanation.
Ben Mildenhall is an American computer scientist known for his work in computer graphics and 3D computer vision, and in particular as the first author of the 2020 paper that introduced neural radiance fields (NeRF), a method for synthesizing photorealistic novel views of a scene from a sparse set of input photographs.[1][2] He was a research scientist at Google Research from 2021 to 2023 and is a co-founder of World Labs, the spatial-intelligence company started in 2024 by Fei-Fei Li and three other researchers.[3][4][5] His work has been recognized with the 2025 ACM Grace Murray Hopper Award and the 2025 SIGGRAPH Significant New Researcher Award, both shared with his frequent collaborator Pratul Srinivasan.[13][14]
Mildenhall's research centers on neural rendering, view synthesis, and inverse graphics, the broad problem of recovering a 3D representation of a scene from 2D images and then rendering new images of it.[2][6] His best-known contribution, NeRF, represents a scene as a continuous volumetric function encoded in the weights of a small neural network; querying that function along camera rays and compositing the results produces images from previously unseen viewpoints.[1] The paper, co-authored with Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and his doctoral advisor Ren Ng, received an Honorable Mention for Best Paper at the 2020 European Conference on Computer Vision (ECCV) and went on to seed a large body of follow-up work in neural scene representation.[1][7]
By mid-2026 his publications had been cited more than fifty thousand times in aggregate, with a Google Scholar h-index of 41; the original NeRF paper alone accounts for roughly twenty thousand of those citations, making it one of the most-cited computer-vision papers of its era.[2][15]
Mildenhall studied at Stanford University, where he earned a bachelor's degree in computer science (with honors) and mathematics, graduating in 2015.[6][8] He attended Stanford from 2011 to 2015 and remained engaged with computer-graphics research throughout, winning the grand prize in the CS348B rendering competition in 2013 and receiving both the Terman Award and the Sterling Award on graduation in 2015.[16] During the summer of 2013 he took part in Stanford's Cross-disciplinary Research in Information Science (CURIS) program, working in Pat Hanrahan's group with graduate students Daniel Ritchie and Matt Fisher on using probabilistic inference for reinforcement learning, work that fed into a 2015 SIGGRAPH paper on controlling procedural modeling programs.[16]
He then entered the doctoral program in electrical engineering and computer sciences at the University of California, Berkeley, completing his PhD in 2020 under the supervision of Ren Ng and supported by a fellowship from the Fannie and John Hertz Foundation, awarded in 2015.[6][2][16] At Berkeley he also held the Tong Leong Lim Pre-Doctoral Prize (2017) and, near the end of his program, the David J. Sakrison Memorial Prize (2021), a departmental award for outstanding doctoral research.[16] His dissertation was titled "Neural Scene Representations for View Synthesis."[7]
Before and during graduate school he held several research internships: at Pixar Animation Studios in 2014, where he worked with Tom Duff, Nelson Max, and Mark Meyer on using sparse voxel octrees to simplify geometry when rendering complex scenes; in Marc Levoy's group at Google Research in 2017, where he worked with Jonathan T. Barron and others on deep learning for multi-image denoising and demosaicking; and at the computer-vision startup Fyusion in 2018, where he worked with Rodrigo Ortiz-Cayon and Abhishek Kar on deep learning for view synthesis.[6][16] The Pixar and Google internships each produced peer-reviewed publications, and the Fyusion collaboration led directly into his early view-synthesis work.[16]
NeRF, short for Neural Radiance Fields, was published in 2020 and is the work most closely associated with Mildenhall.[1] The opening line of the paper summarizes the approach: "We present a method that achieves state-of-the-art results for synthesizing novel views of complex scenes by optimizing an underlying continuous volumetric scene function using a sparse set of input views."[1] Rather than storing explicit geometry such as a mesh or point cloud, the method optimizes a multilayer perceptron that maps a 3D location and a viewing direction to a color and a volume density, which are then integrated along rays using classical volume rendering.[1]
The paper became one of the most-cited works in recent computer graphics, accumulating tens of thousands of citations and prompting an extensive line of research on speeding up, generalizing, and editing radiance fields.[2][9] It is often discussed alongside, and was later partly succeeded in practice by, techniques such as Gaussian splatting for real-time 3D scene reconstruction.
A technical companion to NeRF, "Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains," was published as a spotlight paper at the Conference on Neural Information Processing Systems (NeurIPS) in 2020 and explained why mapping input coordinates through sinusoidal functions allows coordinate-based networks like NeRF to represent fine detail; it has itself been cited several thousand times.[15][16] Mildenhall's earlier 2019 SIGGRAPH paper, "Local Light Field Fusion," developed during his Fyusion internship, established practical guidelines for capturing and interpolating light fields and is widely regarded as a direct precursor to NeRF.[16]
Mildenhall continued to work on neural fields after the original paper. He was a co-author on mip-NeRF (2021), which addressed aliasing in radiance fields and received a Best Paper Honorable Mention at the International Conference on Computer Vision (ICCV), and on its follow-ups mip-NeRF 360 (2022), Block-NeRF (2022), which scaled radiance fields to large outdoor scenes, and Zip-NeRF (2023), an anti-aliased grid-based variant named a Best Paper Finalist at ICCV.[16] He also helped extend radiance fields into generative AI: DreamFusion (2023), built with Ben Poole, Ajay Jain, and Jonathan T. Barron, used a pretrained 2D diffusion model to optimize a NeRF from a text prompt without any 3D training data and received an Outstanding Paper Award at the 2023 International Conference on Learning Representations (ICLR).[16][2] Related lines of work included Dream Fields (2022), an earlier text-to-3D method built with Ajay Jain, Jonathan T. Barron, Pieter Abbeel, and Ben Poole, ReconFusion (2024), which used diffusion priors to reconstruct scenes from very few images, and a body of computational-imaging research such as Burst Denoising with Kernel Prediction Networks (2018) and the lensless DiffuserCam (2017).[16] His publications have appeared at venues including ECCV, the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), ICCV, NeurIPS, and SIGGRAPH.[6]
| Year | Paper | Venue | Note |
|---|---|---|---|
| 2018 | Burst Denoising with Kernel Prediction Networks | CVPR (spotlight) | Multi-frame image denoising[16] |
| 2019 | Local Light Field Fusion | SIGGRAPH | Practical view synthesis, NeRF precursor[16] |
| 2020 | NeRF: Representing Scenes as Neural Radiance Fields | ECCV | Best Paper Honorable Mention; introduced NeRF[1][16] |
| 2020 | Fourier Features Let Networks Learn High Frequency Functions | NeurIPS (spotlight) | Positional encoding for coordinate networks[16] |
| 2021 | Mip-NeRF | ICCV | Best Paper Honorable Mention; anti-aliasing[16] |
| 2022 | Mip-NeRF 360 | CVPR (oral) | Unbounded scenes[16] |
| 2022 | Block-NeRF | CVPR (oral) | Large-scale outdoor scenes[16] |
| 2022 | Ref-NeRF | CVPR | Best Student Paper Honorable Mention[16] |
| 2023 | DreamFusion | ICLR | Outstanding Paper Award; text-to-3D[16] |
| 2023 | Zip-NeRF | ICCV | Best Paper Finalist; grid-based anti-aliasing[16] |
| 2024 | ReconFusion | CVPR | 3D reconstruction with diffusion priors[16] |
After completing his PhD, Mildenhall joined Google Research as a research scientist, a position he held from January 2021 to December 2023.[6][10][16] At Google he worked in David Salesin's group on problems in 3D neural rendering, contributing to the research program that grew out of NeRF and collaborating closely with Jonathan T. Barron, Pratul Srinivasan, Peter Hedman, and Dor Verbin on the mip-NeRF, Block-NeRF, and Zip-NeRF lines of work.[5][4][16] During this period he also served the research community as an area chair for CVPR in 2023 and 2024 and received an outstanding-reviewer award at ECCV 2022.[16]
In 2024 he left to help found World Labs.[10][3] By his own description, his work continues to focus on graphics and 3D computer vision.[6] His longtime co-author Pratul Srinivasan remained on the Google side of the field and is, as of 2026, a research scientist at Google DeepMind.[13]
World Labs is a spatial-intelligence startup founded in 2024 to build large-scale "world models" that can perceive, generate, and interact with 3D environments.[3][11] According to the company, it was established by four "world-renowned" technologists in machine learning, generative AI, computer vision, and graphics: Fei-Fei Li, Justin Johnson, Christoph Lassner, and Mildenhall.[3] Li, sometimes called a "godmother of AI" and creator of the ImageNet dataset, serves as chief executive.[11][4] Mildenhall describes his role at the company simply as building foundation models for spatial intelligence.[16]
The four co-founders and their reported backgrounds are summarized below.
| Co-founder | Background and prior work |
|---|---|
| Fei-Fei Li | Stanford computer-vision researcher; creator of the ImageNet dataset; CEO of World Labs.[11][4] |
| Justin Johnson | Machine-learning researcher known for work on style transfer and vision-language models.[3][4] |
| Christoph Lassner | Computer-vision researcher focused on graphics, data, and 3D reconstruction.[3][4] |
| Ben Mildenhall | Creator of NeRF and other neural-rendering methods; brought 3D-rendering expertise from Google Research.[4][10] |
Investors and the company have repeatedly cited Mildenhall's NeRF work in describing the founding team; the venture firm Andreessen Horowitz, an early backer, called him "the creator of neural radiance fields (NeRF) and many other key generative graphics breakthroughs."[4] In 2024 the company raised financing that valued it at more than $1 billion, in a round whose backers included Andreessen Horowitz, Radical Ventures, NVIDIA, and AMD.[11][12][17]
In November 2025 World Labs released Marble, described as its first commercial product and a "multimodal world model" that generates persistent, editable 3D worlds from inputs such as text, a single image, video, or a coarse 3D layout.[12][18] Marble became generally available on November 12, 2025, after a beta period; the worlds it produces can be explored, expanded, and combined, and exported as Gaussian splats, meshes, or videos, with an experimental editor called Chisel for blocking out spatial layouts before applying a prompted visual style.[18][19] The launch positioned World Labs as a direct competitor in the emerging "world model" segment, where firms aim to generate explorable 3D environments rather than single images or video clips.[18]
In 2022 the Association for Computing Machinery announced that Mildenhall and Pratul Srinivasan had jointly received an Honorable Mention for the 2021 ACM Doctoral Dissertation Award, recognizing their co-invention of the neural radiance field representation, its associated algorithms and theory, and its application to the view-synthesis problem.[7][9] The original NeRF paper itself received a Best Paper Honorable Mention at ECCV 2020.[1][6]
The same partnership earned two further major honors in 2025. At SIGGRAPH 2025 the two received the Significant New Researcher Award for "outstanding contributions to new representations for 3D graphics, neural rendering, novel view synthesis, and generative models of 3D scenes."[14] The award citation credited Mildenhall with combining volume rendering and neural networks into a single scene representation, developing Fourier features and the scaled NeRF variants, and pioneering 3D generative AI through DreamFusion, as well as earlier computational-imaging work on lensless cameras.[14]
In May 2026 the ACM named Mildenhall and Srinivasan recipients of the 2025 ACM Grace Murray Hopper Award, given to an outstanding young computer professional aged 35 or under, "for contributions to radiance field representations, 3D scene capture and rendering, and pioneering neural implicit representations and 3D generative AI."[13][20] The accompanying statement noted that their work had replaced decades of reliance on explicit geometric representations with differentiable neural scene representations and now underpins deployed systems in immersive mapping, 3D commerce, and large-scale scene visualization, with influence extending into medical imaging, astronomy, and computational physics.[13] The Hopper Award carries a prize of $35,000 and is sponsored by Microsoft.[13]
His other distinctions include a Best Paper Honorable Mention at ICCV 2021 for mip-NeRF, a Best Student Paper Honorable Mention at CVPR 2022 for Ref-NeRF, a Best Paper Finalist designation at ICCV 2023 for Zip-NeRF, the David J. Sakrison Memorial Prize (2021), an Outstanding Graduate Student Instructor Award (2021), and selection of his NeRF work as a Communications of the ACM Research Highlight in 2022.[16]