About Me
I am a staff research scientist and tech lead manager at Meta Reality Labs Research, leading a team of AI research scientists focused on advancing 3D computer vision, graphics, and imaging for immersive XR technologies. Before joining Meta in 2017, I earned my Ph.D. in Computer Science with a focus on computational imaging from the University of British Columbia.
My recent research focused on neural rendering for real-time graphics, and 3D computer vision including view synthesis, immersive videos, telepresence, 3D/4D reconstruction and generative content. My work has been published in top-tier venues, including SIGGRAPH, SIGGRAPH Asia, CVPR, ICCV, ECCV, NeurIPS, etc, and has been featured in the Meta Connect keynote, Oculus Blog, demos to Meta CTO and various media outlets. I have also had the privilege of giving invited talks at NVIDIA GTC, Stanford SCIEN talk series, and the CV4MR workshop at CVPR. Please see my publications below for details.
Beyond core algorithm research, as a tech lead, I collaborated with a multidisciplinary team of software engineers, hardware engineers and technical artists, to design and develop real-time, end-to-end prototype demonstrations. Our published work includes gaze-contingent rendering for varifocal VR headsets, real-time perspective-correct MR passthrough, real-time supersampling for high-resolution VR, and ultra-wide field-of-view MR passthrough.
Before I joined Meta, during my PhD study under supervision of Prof. Wolfgang Heidrich at UBC, I worked on computational photography and imaging, including image restoration, superresolution, time-of-flight imaging, non-line-of-sight imaging, etc.
Selected Publications
Geometry-guided Online 3D Video Synthesis with Multi-View Temporal Consistency
Online novel view synthesis of dynamic scenes from multi-view capture, a step towards real-time telepresence
LIRM: Large Inverse Rendering Model for Progressive Reconstruction of Shape, Materials and View-dependent Radiance Fields
Large model for high-quality inverse rendering
Wide Field-of-View Mixed Reality
MR passthrough headset prototype with an ultra-wide field of view — approaching the limits of human vision — designed to create a seamless, almost invisible headset experience
Under Single-Blind Review for SIGGRAPH 2025 Emerging Technologies | Media Report on a previous iteration of our prototype that Meta CTO shared publicly
ReplaceAnything3D: Text-Guided 3D Scene Editing with Compositional Neural Radiance Fields
Text-guided, localized editing method that enables object replacement in a 3D scene
GauFRe: Gaussian Deformation Fields for Real-time Dynamic Novel View Synthesis
View synthesis of dynamic scenes from monocular videos using learned deformable 3D Gaussian Splatting
TextureDreamer: Image-guided Texture Synthesis through Geometry-aware Diffusion
Transfers photorealistic, high-fidelity, and geometry-aware textures from sparse-view images to arbitrary 3D meshes
AlteredAvatar: Stylizing Dynamic 3D Avatars with Fast Style Adaptation
Stylize a dynamic 3D avatar fast
Tiled Multiplane Images for Practical 3D Photography
Efficient 3D photography using tiled multi-plane-images
Temporally-Consistent Online Depth Estimation Using Point-Based Fusion
Online video depth estimation of dynamic scenes via a global point cloud and image-space fusion
NeuralPassthrough: Learned Real-Time View Synthesis for VR
The first learned approach to address the passthrough problem, achieving superior image quality while meeting strict VR requirements for real-time, perspective-correct stereoscopic view synthesis
SNeRF: Stylized Neural Implicit Representations for 3D Scenes
Transfer a neural radiance field to a user-defined style with cross-view consistency
Neural Compression for Hologram Images and Videos
Effective compression method for holograms
Deep 3D Mask Volume for View Synthesis of Dynamic Scenes
High-quality view synthesis of dynamic scenes for immersive videos
Neural Supersampling for Real-time Rendering
The first neural rendering technique for high-fidelity and temporally stable upsampling of rendered content in real-time applications, even in the highly challenging 16x upsampling scenario
DeepFocus: Learned Image Synthesis for Computational Displays
The first real-time neural rendering technique to synthesize physically-accurate defocus blur, focal stacks, multilayer decompositions, and light field imagery using only commonly available RGB-D images
SIGGRAPH ASIA 2018 | Oculus Blog| Oculus Connect Keynote | Media Report
Discriminative Transfer Learning for General Image Restoration
Learning High-Order Filters for Efficient Blind Deconvolution of Document Photographs
Defocus Deblurring and Superresolution for Time-of-Flight Depth Cameras
Stochastic Blind Motion Deblurring
Imaging in Scattering Media Using Correlation Image Sensors and Sparse Convolutional Coding
Temporal Frequency Probing for 5D Transient Analysis of Global Light Transport
Diffuse Mirrors: 3D Reconstruction from Diffuse Indirect Illumination Using Inexpensive Time-of-Flight Sensors
Compressive Rendering of Multidimensional Scenes
Video Processing and Computational Video, LNCS 7082, Springer, 2011