Brian Chao

I'm a final-year PhD candidate at Stanford University working with Prof. Gordon Wetzstein. I'm grateful to be supported by the NSF GRFP and the Stanford Graduate Fellowship.

I study how visual generative models should represent information and allocate computation. My recent work explores adaptive representations for diffusion models across space, scale, and time (Foveated Diffusion, Spectral Progressive Diffusion), with broader interests in video, world models, and the interface between reasoning and visual generation. I've also worked on 4D world modeling at Waymo Research.

Previously, I've worked on efficient representations and rendering for 3D scenes (Textured Gaussians, GWS) and computational imaging systems, including work at Meta Reality Labs.

Email  /  CV  /  Google Scholar  /  Twitter  /  LinkedIn  /  Github

Research

A plug-and-play framework for efficient image and video generation acceleration, using the intrinsic diffusion property of spectral autoregression.

A biologically-inspired diffusion framework that employs spatially adaptive tokenization to concentrate compute on selected regions, achieving up to 4× speedups in image and video synthesis.

An efficient rendering algorithm for multiplane images (MPIs) that enables fast 3D hologram generation for next-generation holographic displays.

A fundamentally new alpha blending algorithm for Gaussian splats generates random-phase, full-bandwidth light field holograms for VR displays, enabling natural defocus blur, physically-accurate parallax, and occlusion.

When state-of-the-art neural rendering meets next-generation holographic VR displays: converting optimized Gaussian splats to holograms that support natural focus cues.

Augmenting Gaussians with RGBA texture maps greatly improves novel-view synthesis quality in 3D Gaussian Splatting.

The novel combination of a multisource laser array and a dynamic Fourier amplitude modulator significantly improves holographic VR display étendue and light field hologram image quality.

A near-eye display design that pairs inverse-designed metasurface waveguides with AI-driven holographic displays to enable full-colour 3D augmented reality from a compact glasses-like form factor.

The inclusion of parallax cues in CGH rendering plays a crucial role in enhancing perceptual realism, and we show this through a live demonstration of 4D light field holograms.

Live demonstration of state-of-the-art image quality, full-color, 3D, holographic VR near-eye displays that supports natural accommodation.

A novel light-efficiency loss function, AI-driven CGH techniques, and camera-in-the-loop calibration greatly improves holographic projector brightness and image quality.

Using coded apertures and a learning-based light field display optimization pipeline to reproduce correct defocus blurs for better accommodation cues.

Leveraging geometric consistency in stereo images to synthesize high-quality light fields.

We propose an image-to-image translation algorithm based on generative adversarial networks that rectifies fisheye images without the need of paired training data.

Using optimal transport to infer regulatory relationships predictive of cellular state changes in single-cell temporal and multimodal data.

Industry Experiences

Working on 4D world modeling and video models with Titas Anciukevičius and Max Jiang, in collaboration with Vincent Casser, Philipp Henzler, and Thomas Funkhouser.

Website source code from here.