Bio

I am currently a PhD candidate at the Center for Visual Computing at UC San Diego, advised by professor Ravi Ramamoorthi and also working closely with professor Tzu-Mao Li. My research in computer graphics primarily focuses on physically based rendering and neural rendering, especially forward&inverse light transport simulation and sampling problems. Previously, I obtained my bachelor's degree in CS from the University of Hong Kong and worked on knowledge-based question answering systems with professor Ben Kao.

I have had the fortune to intern at a few top teams for the past four years: Adobe Graphics Research (2022, with Fujun Luan, Miloš Hašan and Iliyan Georgiev), Meta Reality Labs (2023, with Carl Marshall) and Nvidia Real Time Rendering team (2024-2025, with Marco Salvi), working with amazing mentors and colleagues. Most recently, I'm working with Brian Budge at Meta Reality Labs again.

Before PhD, I worked as a rendering research engineer at ZJU-Manycore Tech joint Lab of CG&AI, led by Dr. Rui Tang (who guided me into graphics research) and Prof. Rui Wang, where I mainly worked on our self-developed Render Engine FF (aka fast and furious :P), Monte Carlo sampling, denoising and other graphics related topics. During the interest exploration phase of my undergraduate, I spent half a year in freezing Ottawa, Canada, working on unified model language, and also spent a short time at Alibaba Group and Sensetime on stream processing and face recognition, respectively.

Research

I am mainly interested in physically-based rendering, neural rendering and real-time graphics, especially light transport simulation, stochastic sampling, learning approaches for appearance modelling and data-prior boosted light transport in general. Recently I'm very interested in optimization problems and 3D geometric learning.

A Generalizable Light Transport 3D Embedding for Global Illumination
Bing Xu, Mukund Varma T, Cheng Wang, Tzu-Mao Li, Lifan Wu, Bart Wronski, Ravi Ramamoorthi, Marco Salvi
Soon to be out
project page / paper / code

We introduce a transformer-based framework for learning a generalizable 3D light transport embedding that directly approximates global illumination from 3D scene geometry, material properties and lighting, without rasterized or path-traced illumination cues. By capturing long-range spatial interactions through attention, our method produces view-independent results across a wide variety of indoor scenes without requiring per-scene training.

BSDF Importance Sampling using a Diffusion Model
Ziyang Fu, Yash Belhe, Haolin Lu, Liwen Wu, Bing Xu, Tzu-Mao Li
ACM SIGGRAPH Asia, 2024
project page / paper / code

BSDF importance sampling, using compact diffusion models to represent spherical domain probability density distributions.

Residual Path Integrals for Re-Rendering
Bing Xu, Tzu-Mao Li, Iliyan Georgiev, Trevor Hedstrom, Ravi Ramamoorthi
Computer Graphics Forum (Proceedings of Eurographics Symposium on Rendering), 2024
Best Paper Award
project page / paper / slides / code (coming)

Incremental re-rendering of scenes in a common senario where only a small portion is moving or edited (e.g. the main character hanging around in an open world). Formulating the difference between two frames as a (correlated) light-transport integral. Devising sampling strategies to focus on paths with non-zero residual-radiance contribution and tailoring appropriate path mappings.
We welcome increased attention to the exploration and potential optimization of the residual path integral and, overall, the re-rendering problem. It is by essence a finite difference renderer and I'm also interested in exploring its applications in inverse rendering.

Importance Sampling BRDF Derivatives
Yash Belhe, Bing Xu, Sai Praveen Bangaru, Ravi Ramamoorthi, Tzu-Mao Li
ACM Transactions on Graphics (TOG), 2024
project page / paper

Importance sampling real-valued derivates, enabling better recovery of spatially varying textures through GD-based inverse rendering.

Neusample: Importance Sampling for Neural Materials
Bing Xu, Liwen Wu, Miloš Hašan, Fujun Luan, Iliyan Georgiev, Zexiang Xu, Ravi Ramamoorthi
ACM SIGGRAPH, 2023
project page / paper / code

A simple toolbox composed of three approaches to efficiently importance sample neural SVBRDFs, potentially facilitating the broader adoption of neural material models in production.

Adversarial Monte Carlo Denoising with Conditioned Auxiliary Feature Modulation
Bing Xu, Junfei Zhang, Rui Wang, Kun Xu, Yong-liang Yang, Chuan Li, Rui Tang
ACM Transactions on Graphics (TOG), ACM SIGGRAPH Asia, 2019
project page / paper / code

Denoising Monte Carlo renderings by better utilizing the first bounce auxiliary features and adversarial loss.

Gallery

If it can be called A Gallery...., I do some drawing and painting when I have spare time after paper deadlines or when I am stuck with project. I might stop doing research once my paintings start making money - you know, the best-case scenario I can think of, after the painter's death.

These photos were taken randomly so the resolution may not be good. If you are up for charity, these are available for $2000 per piece :)

I have always been a visual person and am born sensitive to colors. Expressing and communicating via pixels are my choice of words. Painting is one form, ray tracing is another. I find beauty in screen language and passion in filmmaking.

Miscellany

I play some soccer with our community amateur team when the sun is not too bright (what does this mean in San Diego?).