I am a Ph.D candidate at the University of California, Santa Barbara (UCSB). My research interests are physically-based rendering, realtime rendering, with appearance modeling and synthesis. I love animations, photography, and video games.
Email: yaoyibai@ucsb.edu
Education
University of California, Santa Barbara, — Ph.D. in Computer Science, 2019.09 — current
Ph.D. Dissertation: Elegant Precomputation in Rendering
University of Pennsylvania, Computer Graphics and Game Technology, Master’s Degree, 2016 - 2018
Wuhan University, Simulation Science and Technology, Bachelor’s Degree, 2012 - 2016
Work Experience
Innopeak Technology Inc. — Research Intern, 2023.06 — 2023.09
Tencent America — Research and Development Intern, 2022.06 — 2023.01
Facebook Reality Labs — Research Intern (FRL Research), 2021.06 — 2021.09
Adobe Inc. — Emerging Graphics Group Intern, 2020.06 — 2020.09
Venturi Labs — Software Developer Internship, 2018.08 — 2019.05
Publications
BSDF Importance Baking: A Lightweight Neural Solution to Importance Sampling Parametric BSDFs — arXiv preprint arXiv:2210.13681, 2022 - arxiv.org
Yaoyi Bai, Songyin Wu, Zheng Zeng, Beibei Wang, Ling-Qi Yan
In this paper, we seek an efficient and general way for importance sampling parametric BRDFs. We notice that the nature of importance sampling is the mapping between a uniform distribution and the target distribution. Specifically, when BRDF parameters are given, the mapping that performs importance sampling on a BRDF slice can be simply recorded as a 2D image that we name as importance map. Following this observation, we accurately precompute the importance maps using a mathematical tool named optimal transport. Then we propose a lightweight neural network to efficiently compress the precomputed importance maps. In this way, we have brought parametric BRDF important sampling to the precomputation stage, avoiding heavy runtime computation. Since this process is similar to light baking where a set of images are precomputed, we name our method importance baking. Together with a BRDF evaluation network and a PDF (probability density function) query network, our method enables full multiple importance sampling (MIS) without any revision to the rendering pipeline.
Neural Complex Luminaires: Representation and Rendering — ACM Transactions on Graphics (Proceedings of SIGGRAPH 2021)
Junqiu Zhu, Yaoyi Bai (main contributor), Zilin Xu (main contributor), Steve Bako, Edgar Velázquez-Armendáriz, Lu Wang, Pradeep Sen, Miloš Hašan, Ling-Qi Yan
Complex luminaires, such as grand chandeliers, can be extremely costly to render because the light-emitting sources are typically encased in complex refractive geometry, creating difficult light paths that require many samples to evaluate with Monte Carlo approaches. Our work can easily plug into conventional renderers, as it works with the standard techniques of path tracing and multiple importance sampling (MIS). Our solution is to train three networks to perform the essential operations for evaluating the complex luminaire at a specific point and view direction, importance sampling a point on the luminaire given a shading location, and blending to determine the transparency of luminaire queries to properly composite them with other scene elements. We perform favorably relative to state-of-the-art approaches and render final images that are close to the high-sample-count reference with only a fraction of the computation and storage costs, with no need to store the original luminaire geometry and materials.
Precomputed Dynamic Appearance Synthesis and Rendering (EGSR 2024)
Yaoyi Bai, Miloš Hašan, Ling-Qi Yan
Interpolation between objects of varying dimensionality is a common task in computer graphics; however, high-quality dynamic natural interpolation for appearance remains scarce. In this paper, we propose a blending framework for general appearances that can be integrated into renderers without modifying the rendering pipeline. For natural interpolation calculations, we use the mathematical tool optimal transport (OT), known for its promising blending quality. Although recent advancements in OT theory have improved computational performance, integrating runtime OT calculations into the path tracing rendering pipeline compromises algorithm efficiency and increases storage requirements. To address this, we propose a novel solution that precomputes appearances into a proxy distribution and introduces a hierarchical query structure. This enables efficient online point or range data querying, allowing for the generation or retrieval of large data sets as needed. Additionally, the proxy and hierarchical query structure facilitate multi-way barycenter computation. With this efficient query structure and barycentric calculation, we demonstrate several applications of our method, including 2D and 3D interpolation, as well as isotropic BRDF interpolation.
Patent
Methods, systems, and computer readable media for extended reality user interface — University of Pennsylvania
Stephen H Lane, Matthew Anthony Boyd-Surka, Yaoyi Bai, Aline Sarah Normoyle
SIGGRAPH 2020 Appy Hour Talk, University of Pennsylvania
Bubble app enables an augmented reality (AR) user interface to be created as a hierarchy of interactable nested bubbles. It allows users to easily navigate, explore, and interact with content and other information (text, images, videos, 3D models) in a manner naturally suited to mixed reality and AR worlds.
Other Projects
Real-time Human Hair Rendering Optimization using OpenGL — Tencent America
Advisor: Dr. Kui Wu
Human hair rendering in video games presents several challenges. Firstly, hair strands often display aliasing when viewed from a distance, as they are much thinner than a pixel. Similar issues arise with hair shadows. Additionally, the sheer number of hair strands per character makes enabling global illumination on desktop platforms time-consuming. In this project, we developed a human hair rendering framework using OpenGL, incorporating the Kajiya-Kay model, the Marschner model, shadow maps, deep opacity maps, and dual scattering techniques. We then explored further optimizations for software anti-aliasing.
By Example Appearance Synthesis with Optimal Transport — Facebook Reality Labs
Advisor: Dr. Zhao Dong
We are dedicated to exploring a rapid and lifelike by-example appearance synthesis solution suitable for straightforward appearances across various dimensions. Our chosen synthesis tool is optimal transport (OT), renowned for its ability to achieve natural interpolation between distributions. Beginning with 2D textures and 3D point clouds, we have achieved impressive by-example blending results for these elements.
Simulating Elastic Materials with MPM Solver
Contributors: Joshuah Wolper, Ziyin Qu, Yaoyi Bai. Advisors: Prof. Chenfanfu Jiang, Dr. Andre Pradhana.
This project is modeling elastic materials both in freefall and collision using a Material Point Method (MPM) solver.
Cesium Snow Rendering
Contributors: Yaoyi Bai, Yuxin Hu, Dr. Anton Khabbaz. Advisors: Patrick Cozzi, Gary Kangning Li.
Cesium has great Terrain system that is capable of displaying accurate shapes of mountains and glaciers. Cesium displays high or steep terrains with perspective, allowing the user to physically see mountain heights and shadows. Before, snow is added as a texture, which has limitations: the texture modifies a preexisting scene to make it snow covered, and offers no depth information. We implemented the add in real-time snow on the terrain would make the system more realistic. Our project also enabled snow accumulation in real-time through Perlin noise. Generally, we focused on rendering snow with the idea of creating a framework to incorporate in the future other weather-related changes in real time.
Magic Offspring Maya Plugin
Contributors: Yaoyi Bai, Yuxin Hu. Advisor: Prof. Stephen H Lane.
This project implemented a shape blending Maya plugin based on the paper Topology-Varying 3D Shape Creation via Structural Blending written by Alhashim, I. Magical Offspring is a tool for fast generation of novel shapes by blending a source and target shape of a same object category, but with different topological and geometric structure. It helps game developers generate many objects like furniture, robotics, airplanes with variant shapes that are usually needed in games, without requiring developers to manually model every shape themselves.