Ben Mildenhall

I am a research scientist at Google Research, where I work on problems in computer vision and graphics. I recently received my PhD from UC Berkeley, where I was advised by Ren Ng and supported by a Hertz fellowship.

In the summer of 2017, I was an intern in Marc Levoy's group in Google Research. In the summer of 2018, I worked with Rodrigo Ortiz-Cayon and Abhishek Kar at Fyusion. I did my undergrad at Stanford University and worked at Pixar Research in the summer of 2014.

Email  /  CV  /  Google Scholar  /  Twitter  /  GitHub

We train RawNeRF directly on linear raw camera images, enabling new HDR view synthesis applications and greatly increasing robustness to camera noise.

We extend mip-NeRF to produce photorealistic results on unbounded scenes.

Explicitly modeling reflections in NeRF produces realistic shiny surfaces and accurate surface normals, and lets you edit materials.

We build city-scale scenes from many NeRFs, trained using millions of images.

We regularize unseen views during optimization to enable view synthesis from as few as 3 input images.

Supervising the CLIP embeddings of NeRF renderings allows us to generate 3D objects from text prompts alone.

We apply dense depth completion techniques to freely-available sparse stereo data to guide NeRF reconstructions from few input images.

We denoise images efficiently by predicting spatially-varying kernels at low resolution and using a fast fused op to jointly upsample and apply these kernels at full resolution.

We prefilter the positional encoding function and train NeRF to generate anti-aliased renderings.

We bake a trained NeRF into a sparse voxel grid of colors and features in order to render it in real-time.

We use meta-learning to find weight initializations for coordinate-based MLPs that allow them to converge faster and generalize better.

We recover relightable NeRF-like models using neural approximations of expensive visibility integrals, so we can simulate complex volumetric light transport during training.

We demonstrate that composing fully-connected networks with a simple Fourier feature mapping allows them to learn much high frequency functions.

We recover relightable NeRF-like models by predicting per-location BRDFs and surface normals, and marching light rays through the NeRV volume to compute visibility.

We optimize a simple fully-connected network to represent a single scene as a volume, then use volume rendering to do view synthesis.

We represent scenes as multi-layer panoramas with depth for VR view synthesis.

We predict a volume from an input stereo pair that can be used to calculate incident lighting at any 3D point within a scene.

We can hide hyperlinks in natural images to create aesthetically pleasing barcodes.

We develop and analyze a deep learning method for rendering novel views of complex real world scenes.

We can learn a better denoising model by processing and unprocessing images the same way a camera does.

Burst Denoising with Kernel Prediction Networks
Ben Mildenhall, Jonathan T. Barron, Jiawen Chen, Dillon Sharlet, Ren Ng, Robert Carroll
Computer Vision and Pattern Recognition (CVPR), 2018 (spotlight)
project page / arXiv / code

We train a network to predict linear kernels that denoise bursts of raw linear images.

DiffuserCam: Lensless Single-exposure 3D Imaging
Nick Antipa, Grace Kuo, Reinhard Heckel, Ben Mildenhall, Emrah Bostan, Ren Ng, Laura Waller
Optica, 2018
project page / arXiv

Using a diffuser instead of a lens lets you recover 3D in a single exposure.

Approximations for the distribution of microflake normals
Nelson Max, Tom Duff, Ben Mildenhall, Yajie Yan
The Visual Computer, 2017

We precompute microflake approximations to make rendering large meshes at a distance more efficient.

Controlling Procedural Modeling Programs with Stochastically-Ordered Sequential Monte Carlo
Daniel Ritchie, Ben Mildenhall, Noah D. Goodman, Pat Hanrahan
SIGGRAPH, 2015

We improve control over the output of highly-variable procedural modeling programs by using SOSMC to provide incremental feedback on partially-generated models.

Extending the PBRT Renderer to Support Volumetric Light Sources
Grand prize in CS384B rendering competition, Spring 2013

Adding support for multicolored, nonhomogeneous, emissive volumes in PBRT 2's path tracing integrator.

Ray Tracer
CS148 assignment, Fall 2012

Implemented a ray tracer with reflections, refractions, soft shadows, and depth of field.

CS184 - Summer 2020 (Co-instructor)

CS184 - Spring 2017 (GSI)

CS184 - Spring 2016 (GSI)

Rachmaninoff - Etude-Tableau Op. 39 No. 5
Encore after a salon recital, Spring 2019

Rachmaninoff - Rhapsody on a Theme of Paganini
With Stanford Symphony Orchestra, Spring 2015

Brahms - Piano Quintet
Recital at Stanford, Winter 2015

Yep it's another Jon Barron website.
Last updated March 2022.