About Me
I am a Machine Learning Engineer on the Avatar AI team at Roblox, where I work on 3D generative AI for avatars, including avatar generation, rigging, and animation.
Previously, at Meta Reality Labs, I worked on automatic 3D face rigging and on audiovisual motion models and video generation for real-time interactive avatars. Before that, in the Intelligent Creation Lab at ByteDance / TikTok, I worked on portrait animation, talking-head and human video generation, and real-time facial animation for TikTok Avatars. Earlier, I worked on large-scale 3D scene reconstruction at Matterport.
I received my Ph.D. in Computer Science from the University of Texas at Dallas, advised by Prof. Xiaohu Guo, and my M.S. and B.S. from Tsinghua University, where I was advised by Prof. Yu-shen Liu and Prof. Jun-hai Yong.
News
- Oct 2026Our team launched Prompt-to-Avatar in Roblox Studio Beta, generating animatable 3D game characters with expressive dynamic heads from text prompts. New
- Aug 2026OmniFaceRig is accepted to SIGGRAPH Asia 2026 (ACM TOG), and its dataset of 1,345 rigged 3D faces is released. New
- Jun 2025X-Dancer is accepted to ICCV 2025 as a Highlight.
- Jun 2025Meta AI released Seamless Interaction, a family of dyadic audiovisual motion models for interactive avatars, together with a large-scale conversational dataset (blog).
- Feb 2025X-Dyna is accepted to CVPR 2025 as a Highlight.
- Sep 2024Our real-time interactive 2D avatars, embodied AI agents with expressive faces and accurate lip sync, are showcased in the Meta Connect 2024 keynote.
- Apr 2024X-Portrait is accepted to SIGGRAPH 2024.
- Jun 2022Our team launched TikTok Avatars, real-time face-driven 3D avatars used in 1M+ videos with 50M+ plays.
Projects
-
Roblox · 2026A Roblox Studio feature, built by the Avatar AI team, that generates rigged and animatable 3D avatars from text prompts, with optional dynamic heads and expressive faces, for creators and game builders. Available in Studio Beta, Assistant, and Studio MCP. -
Meta Reality Labs · 2025Contributed to dyadic audiovisual motion models that generate the facial expressions and body gestures of two people in face-to-face conversation, driving embodied AI agents as both 2D video and 3D avatars. Released by Meta AI with a large-scale conversational dataset. -
Meta Reality Labs · 2024Co-developed real-time facial animation for interactive 2D avatars, embodied AI agents that hold LLM-driven conversations with expressive emotion, natural head motion, and accurate lip sync. Showcased in the Meta Connect 2024 keynote. -
ByteDance / TikTok · 2022Co-developed the real-time facial animation for TikTok Avatars, a lightweight video-to-blendshape model running on mobile. Used in 1M+ videos with 50M+ plays.
Publications
-
SIGGRAPH Asia 2026 (ACM TOG) -
Arxiv Preprint, 2025/07 Technical Report -
The International Conference on Computer Vision (ICCV), 2025 ICCV Highlight -
IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR), 2025 CVPR Highlight -
Computational Visual Media (CVM), 2025 Oral -
-
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024 Poster -
AI-Generated Characters: Putting Deepfakes to Good Use (CHI Workshop), 2022 Oral -
Arxiv Preprint, 2022 -
-
-
Computational Visual Media (CVM), 2017 Oral -
-
Computer-Aided Design (CAD), 2012
Work Experience
-
Roblox Machine Learning Engineer, Avatar AI2026 – Present -
Meta Reality Labs AI Research Engineer, XR-Tech 3D GenAI2024 – 2026 -
ByteDance / TikTok AI Research Engineer, Intelligent Creation Lab2020 – 2024 -
Matterport Computer Vision Engineer2018 – 2020
Education
-
The University of Texas at Dallas Ph.D. in Computer Science2018 -
Tsinghua University M.S. in Computer Science2012 -
Tsinghua University B.S. in Automation2009 -
Peking University B.S. Minor in Economics2011
Services
- Conference Reviewer
- CVPR, ECCV, ICCV, AAAI, NeurIPS, SIGGRAPH, SIGGRAPH Asia, ACM MM, WACV, GMP, PacificVis, SPM, CVM, etc.
- Journal Reviewer
- TOG, TVCG, TMM, GMOD, CASA, CAD, CAGD, TVCJ, CAVW, GM, WEVJ, Measurement Science and Technology (MST), Remote Sensing, Sensors, Drones, Electronics, Mathematics, Biomimetics, Forests, Symmetry, etc.
Side Projects
-
An automated daily digest of Hacker News best stories, powered by the official HN API and GPT for bilingual (English & Chinese) summaries, with AI-generated briefs, metadata, and preview images — updated every day via GitHub Actions.
-
A Chinese-first technical blog for deep learning study notes, covering mathematical foundations, generative models, VAE/GAN, Transformer architecture, formulas, diagrams, and implementation details.