Zhuo Su | 苏卓

I am a Tech Lead and Researcher at ByteDance, working on Human-Centric AI. My research focuses on enabling machines to perceive, understand, generate and interact with human-centric worlds, bridging the physical and virtual worlds through human capture, avatar systems, motion intelligence, and immersive interaction. My long-term goal is to develop machine intelligence grounded in human experiences, capable of understanding people, modeling human behaviors, and enabling natural interactions between humans, AI agents, robots and worlds. Prior to joining ByteDance, I worked at Tencent as a Senior Researcher through the Special Recruitment Talents Program (技术大咖). Before entering the industry, I graduated from the Department of Automation at Tsinghua University, where I had the honor of being supervised by Qionghai Dai and Lu Fang, and collaborated closely with Yebin Liu and Lan Xu. Email: suzhuo13@gmail.com | suzhuo@bytedance.com Research Applications Invited Talks Awards Google Scholar

• POLAR: A Portrait OLAT Dataset and Generative Framework for Illumination-Aware Face Modeling Zhuo Chen*, Chengqun Yang*， Zhuo Su‡ (Corresponding Author), Zheng Lv, Jingnan Gao, Xiaoyuan Zhang, Xiaokang Yang, Yichao Yan‡ CVPR 2026 Oral We introduce POLAR, a large-scale, physically calibrated OLAT dataset with 200+ subjects captured under 156 lighting directions, multiple views, and diverse expressions. Based on this, we propose POLARNet, a flow-based model that predicts per-light OLAT responses from a single portrait, capturing fine-grained, direction-aware illumination while preserving identity. Paper Project page

• FlexAvatar: Flexible Large Reconstruction Model for Animatable Gaussian Head Avatars with Detailed Deformation Cheng Peng*, Zhuo Su*‡ (Corresponding Author), Liao Wang*, Chen Guo, Zhaohu Li, Chengjiang Long, Zheng Lv, Jingxiang Sun, Chenyangguang Zhang, Yebin Liu‡ CVPR 2026 We present FlexAvatar, a flexible large reconstruction model that generates high fidelity animatable 3D head avatars with detailed dynamic deformations from single or sparse images without requiring camera poses or expression labels, using a transformer based canonical representation together with a UV conditioned lightweight decoder for real time expression aware detail synthesis. Paper Project page

• GeoDiff4D: Geometry-Aware Diffusion for 4D Head Avatar Reconstruction Chao Xu, Xiaochen Zhao, Xiang Deng, Jingxiang Sun, Donglin Di, Zhuo Su†(Project Lead), Yebin Liu‡ CVPR 2026 We propose a portrait video generation framework from a single portrait image, which jointly synthesizes images and surface normals while learning pose-free expression representations for accurate geometry and real-time rendering, with optional integration into a 3D Gaussian-based avatar. Paper Project page

• Relightable and Animatable Gaussian Head Avatar from Monocular Videos Zhuo Chen*, Yichao Yan‡, Jingnan Gao, Zhuo Su†(Project Lead), Chao Wen, Zhaohu Li, Yuhao Cheng, Xueying Lee, Yutong Leng, Yikun Zeng, Guidong Wang, Xiaokang Yang TVCG 2026 We propose a method for relightable avatar reconstruction from a monocular video under unknown lighting, using a disentangled 3D Gaussian representation and a diffusion model to predict canonical lighting and material maps for improved rendering and view consistency. Paper Project page

• Director: Instance-aware Gaussian Splatting for Dynamic Scene Modeling and Understanding Yuheng Jiang, Yiwen Cai, Zihao Wang, Yize Wu, Sicheng Li, Zhuo Su, Shaohui Jiao, Lan Xu arXiv 2026 We propose Director, a unified spatio-temporal Gaussian representation for dynamic scene modeling that jointly enables temporally coherent 4D reconstruction, high-fidelity human rendering, and instance-level semantic understanding through language-aligned semantic supervision, motion refinement, and geometry-aware regularization. Paper Project page

• InterCoser: Interactive 3D Character Creation with Disentangled Fine-Grained Features Yi Wang*, Jian Ma*, Zhuo Su†(Project Lead), Guidong Wang, Jingyu Yang, Yu-Kun Lai, Kun Li‡ AAAI 2026 Oral We present an interactive AIGC framework for disentangled 3D character creation, supporting sketch-based clothing transfer and 3D-proxy-guided part editing. Our method leverages geometry-aware 3D Gaussian reconstruction to ensure high-fidelity, editable avatars with fine-grained control. Paper Project page

• DreamCoser: Controllable Layered 3D Character Generation and Editing Yi Wang*, Jian Ma*, Zhuo Su†(Project Lead), Guidong Wang, Yu-Kun Lai, Kun Li‡ SIGGRAPH Asia 2025 Technical Communications We present DreamCoser, an AIGC framework for controllable, layered 3D character generation and sketch-based editing, enabling part-level customization from a single image. Paper Project page

• Topology-Aware Optimization of Gaussian Primitives for Human-Centric Volumetric Videos Yuheng Jiang, Chengcheng Guo, Yize Wu, Yu Hong, Shengkun Zhu, Zhehao Shen, Yingliang Zhang, Shaohui Jiao , Zhuo Su, Lan Xu, Marc Habermann, Christian Theobalt SIGGRAPH Asia 2025 We propose TaoGS, a framework that disentangles motion and appearance for coherent volumetric video under topological variation, with codec-friendly compression for scalable high-fidelity rendering. Paper Project page

• Parametric Gaussian Human Model: Generalizable Prior for Efficient and Realistic Human Avatar Modeling Cheng Peng, Jingxiang Sun, Yushuo Chen, Zhaoqi Su, Zhuo Su, Yebin Liu 3DV 2026 We present PGHM, a generalizable and efficient framework that integrates human priors into 3DGS for fast and high-fidelity avatar reconstruction from monocular videos. Paper Project page

• SMGDiff: Soccer Motion Generation using Diffusion Probabilistic Models Hongdi Yang, Chengyang Li, Zhenxuan Wu, Gaozheng Li, Jingya Wang, Jingyi Yu, Zhuo Su, Lan Xu ICCV 2025 We introduce SMGDiff, a novel two-stage framework for generating real-time and user-controllable soccer motions. Our key idea is to integrate real-time character control with a powerful diffusion-based generative model, ensuring high-quality and diverse output motion. Paper Project page

• SEGA: Drivable 3D Gaussian Head Avatar from a Single Image Chen Guo*, Zhuo Su*‡ (Corresponding Author), Jian Wang , Shuang Li, Xu Chang, Zhaohu Li, Yang Zhao, Guidong Wang, Ruqi Huang‡ arXiv 2025 We propose SEGA, a novel approach for Single-imagE-based 3D drivable Gaussian head Avatar creation that combines generalized prior models with a new hierarchical UV-space Gaussian Splatting framework. Paper Project page

• EnvPoser: Environment-aware Realistic Human Motion Estimation from Sparse Observations with Uncertainty Modeling Songpengcheng Xia, Yu Zhang, Zhuo Su† (Project Lead), Xiaozheng Zheng, Zheng Lv, Guidong Wang, Yongjie Zhang, Qi Wu, Lei Chu, Ling Pei‡ CVPR 2025 We propose EnvPoser, a two-stage method using sparse tracking signals and pre-scanned environment from VR devices to perform full-body motion estimation and handle the multi-hypothesis nature with uncertainty-aware and environmental constraint integration. Paper Project page

• RePerformer: Immersive Human-centric Volumetric Videos from Playback to Photoreal Reperformance Yuheng Jiang, Zhehao Shen, Chengcheng Guo, Yu Hong, Zhuo Su, Yingliang Zhang, Marc Habermann‡, Lan Xu‡ CVPR 2025 We present RePerformer, a Gaussian-based representation for high-fidelity volumetric video playback and re-performance. Via Morton-based parameterization, our method enables efficient rendering. A semantic-aware alignment module and deformation transfer enhance realistic motion re-performance. Paper Project page

• EMHI: A Multimodal Egocentric Human Motion Dataset with HMD and Body-Worn IMUs Zhen Fan*, Peng Dai*, Zhuo Su*, Xu Gao, Zheng Lv, Jiarui Zhang, Tianyuan Du, Guidong Wang, Yang Zhang AAAI 2025 We introduce EMHI, a dataset combining stereo images from headsets and IMU data for egocentric human motion capture in VR. It includes 28.5 hours of data from 58 subjects. We also propose MEPoser, a method that effectively uses this multimodal data for improved pose estimation. Paper Project page

• HeadGAP: Few-shot 3D Head Avatar via Generalizable Gaussian Priors Xiaozheng Zheng, Chao Wen, Zhaohu Li, Weiyi Zhang, Zhuo Su, Xu Chang, Yang Zhao, Zheng Lv, Xiaoyuan Zhang, Yongjie Zhang, Guidong Wang, Lan Xu 3DV 2025 We propose a 3D head avatar creation method that generalizes from few-shot in-the-wild data. By using 3D head priors from a large-scale dataset and a Gaussian Splatting-based network, our approach achieves high-fidelity rendering and robust animation. Paper Project page

• LoGAvatar: Local Gaussian Splatting for human avatar modeling from monocular video Jinsong Zhang, Xiongzheng Li, Hailong Jia, Jin Li, Zhuo Su, Guidong Wang, Kun Li‡ CAD 2025 We present LoGAvatar, a hierarchical 3D Gaussian Splatting framework that reconstructs editable avatars from monocular videos by anchoring local Gaussians to a human template and using a convolution-based texture atlas for real-time rendering and intuitive editing. Paper Project page

• HumanSplat: Generalizable Single-Image Human Gaussian Splatting with Structure Priors Panwang Pan*, Zhuo Su* † (Project Lead), Chenguo Lin*, Zhen Fan, Yongjie Zhang, Zeming Li, Tingting Shen, Yadong Mu, Yebin Liu‡ NeurIPS 2024 We propose HumanSplat, a method that predicts the 3D Gaussian Splatting properties of a human from a single input image in a generalizable way. It utilizes a 2D multi-view diffusion model and a latent reconstruction transformer with human structure priors to effectively integrate geometric priors and semantic features. Paper Project page

• HiFi4G: High-Fidelity Human Performance Rendering via Compact Gaussian Splatting Yuheng Jiang, Zhehao Shen, Penghao Wang, Zhuo Su, Yu Hong, Yingliang Zhang, Jingyi Yu, Lan Xu CVPR 2024. We present an explicit and compact Gaussian-based approach for high-fidelity human performance rendering from dense footage, in which our core intuition is to marry the 3D Gaussian representation with non-rigid tracking. Paper Project page

• Joint2Human: High-quality 3D Human Generation via Compact Spherical Embedding of 3D Joints Muxin Zhang, Qiao Feng, Zhuo Su, Chao Wen, Zhou Xue, Kun Li CVPR 2024 We introduce Joint2Human, a novel method that leverages 2D diffusion models to generate detailed 3D human geometry directly, ensuring both global structure and local details. Paper Project page

• OHTA: One-shot Hand Avatar via Data-driven Implicit Priors Xiaozheng Zheng, Chao Wen, Zhuo Su, Zeran Xu, Zhaohu Li, Yang Zhao, Zhou Xue CVPR 2024 OHTA is a novel approach capable of creating implicit animatable hand avatars using just a single image. It facilitates 1) text-to-avatar conversion, 2) hand texture and geometry editing, and 3) interpolation and sampling within the latent space. Paper Project page

• HMD-Poser: On-Device Real-time Human Motion Tracking from Scalable Sparse Observations Peng Dai, Yang Zhang, Tao Liu, Zhen Fan, Tianyuan Du, Zhuo Su, Xiaozheng Zheng, Zeming Li CVPR 2024 We propose HMD-Poser, the first unified approach to recover full-body motions using scalable sparse observations from HMD and body-worn IMUs. In particular, it can support a variety of input scenarios, such as HMD, HMD+2IMUs, HMD+3IMUs, etc. Paper Project page

• Realistic Full-Body Tracking from Sparse Observations via Joint-Level Modeling Xiaozheng Zheng*, Zhuo Su*, Chao Wen, Zhou Xue, Xiaojie Jin ICCV 2023 We propose a two-stage framework that can obtain accurate and smooth full-body motions with the three tracking signals of head and hands only, in which we first explicitly model the joint-level features and then utilize them as spatiotemporal transformer tokens to capture joint-level correlations. Paper Project page

• Instant-NVR: Instant Neural Volumetric Rendering for Human-object Interactions from Monocular RGBD Stream Yuheng Jiang, Kaixin Yao, Zhuo Su, Zhehao Shen, Haimin Luo, Lan Xu CVPR 2023 We propose a neural approach for instant volumetric human-object tracking and rendering using a single RGBD camera. It bridges traditional non-rigid tracking with recent instant radiance field techniques via a multi-thread tracking-rendering mechanism. Paper Project page

• Robust Volumetric Performance Reconstruction under Human-object Interactions from Monocular RGBD Stream Zhuo Su, Lan Xu, Dawei Zhong, Zhong Li, Fan Deng, Shuxue Quan, Lu Fang TPAMI 2022 We propose a robust volumetric performance reconstruction system for human-object interaction scenarios using only a single RGBD sensor, which combines various data-driven visual and interaction cues to handle the complex interaction patterns and severe occlusions. Paper Project page

• NeuralHOFusion: Neural Volumetric Rendering Under Human-Object Interactions Yuheng Jiang, Suyi Jiang, Guoxing Sun, Zhuo Su, Kaiwen Guo, Minye Wu, Jingyi Yu, Lan Xu CVPR 2022 We propose a robust neural volumetric rendering method for human-object interaction scenarios using 6 RGBD cameras, achieving layer-wise and photorealistic reconstruction results of human performance in novel views. Paper Project page

• Learning Variational Motion Prior for Video-based Motion Capture Xin Chen*, Zhuo Su*, Lingbo Yang*, Pei Cheng, Lan Xu, Gang Yu arXiv 2022 We propose a novel variational motion prior (VMP) learning approach for video-based motion capture. Specifically, VMP is implemented as a transformer-based variational autoencoder pretrained over large-scale 3D motion data, providing an expressive latent space for human motion at sequence level. Paper Project page

• RobustFusion: Human Volumetric Capture with Data-driven Visual Cues using a RGBD Camera Zhuo Su, Lan Xu, Zerong Zheng, Tao Yu, Yebin Liu, Lu Fang ECCV 2020 (Spotlight) We introduce a robust human volumetric capture approach combined with various data-driven visual cues using a Kinect, which outperforms existing state-of-the-art approaches significantly. Paper Project page

• UnstructuredFusion: Realtime 4D Geometry and Texture Reconstruction using Commercial RGBD Cameras Lan Xu, Zhuo Su, Lei Han, Tao Yu, Yebin Liu, Lu Fang TPAMI 2019 We propose UnstructuredFusion, which allows realtime, high-quality, complete reconstruction of 4D textured models of human performance via only three commercial RGBD cameras. Paper Project page

Applications

Avatar Creation and XR Telepresence

This line of work includes 8 patents, with technology transfer to XR avatar and holographic telepresence products for realistic human representation and interaction in virtual environments.

Motion Capture and Human Motion Understanding

This line of work includes 3 patents, with successful technology transfer to real-world XR products including Pico Motion Tracker, enabling precise and accessible full-body motion tracking in consumer virtual reality.

Public patents include:

• "Method and apparatus for pose estimation, and electronic device", Patent, US20250028385A1

Dynamic Human Reconstruction

This line of work includes 4 patents, with technology transfer to industrial partners and deployment in products such as fitness measurement and related 3D human-centric applications.

Public patents include:

• “Dynamic three-dimensional reconstruction method, device, equipment, medium and system”, Patent, CN:201910110062:A
• “Texture real-time determination method, device and equipment for dynamic scene and medium”, Patent, CN:201910110044:A
• “Dynamic human body three-dimensional reconstruction method, device, equipment and medium”, Patent, CN:202010838902:A
• “Dynamic human body three-dimensional model completion method and device, equipment and medium”, Patent, CN:202010838890:A

Static Reconstruction and SLAM

This line of work includes 3 patents, with technology transfer to industrial partners including smartphone manufacturers, and deployment in navigation, AR effects, and related mobile 3D vision scenarios.

Public patents include:

• “Depth Camera-based three-dimensional renconstruction method and apparatus, device, and storage medium”, Patent, US201916977899A
• “Depth camera calibration method and device, electronic equipment and storage medium”, Patent, CN:201810179738:A
• “A three-dimensional rebuilding method and device based on a depth camera, an apparatus and a storage medium”, Patent, CN:201810179264:A



Invited Talks

Academic Conference

• 

  Human-Centric Capture and Digitalization for Immersive XR Experiences

  This talk presents recent advances in human-centric capture and digitalization for XR, highlighting progress in motion capture, 3D reconstruction, performance capture, and avatar creation. We discuss how these pipelines work together to faithfully represent human motion, appearance, and behavior in virtual environments. By unifying these elements, we show how real people can be faithfully brought into immersive virtual worlds to enable true presence.

  Aug. 13, 2025, SIGGRAPH, Vancouver | Technical Workshop

  Slide Webpage

Industry Forum

• 

  Towards Deployable High-Fidelity Avatars: A 3D Modeling Framework Powered by Scalable Real-World Data

  The deployment of high-fidelity 3D avatars is often hindered by the profound gap between ideal lab conditions and chaotic real-world inputs. This work tackles this "Reality Gap" by exploring a scalable, data-driven modeling framework tailored for practical user scenarios. Spanning from foundational data synthesis and illumination decoupling to flexible Large Reconstruction Models, I share this exploration to demonstrate how we are powering truly deployable digital humans.

  Apr. 17, 2026 China3DV, Hangzhou | Young Scholar Forum

  Slide Webpage
• 

  Human Motion Capture and Avatar Creation using Sparse Observations

  The development of high-fidelity 3D human motion capture and avatar reconstruction is essential for immersive experiences in XR applications. This line of work explores modeling under sparse observation settings, tackling the challenges posed by limited sensors and minimal image inputs. The research spans from multi-modal motion understanding to generalizable avatar generation, and I share this exploration with the hope of contributing to the XR technologies.

  Apr. 11, 2025 China3DV, Beijing | Young Scholar Forum

  WeChat Post Webpage

ByteTech Technical Sharing Series



Revolutionizing XR MoCap: Sparse to Multi-Modal, Environment-Aware

This talk reviews four representative works — AvatarJLM, HMDPoser, EnvPoser, and EMHI — showcasing advances from sparse IMU modeling and egocentric sensing to multi-modal fusion and environment-aware inference, pushing sparse motion capture toward practical XR applications.

Jun. 5, 2025, ByteDance, Online Live Stream

Slide WeChat Post



Human 3D Reconstruction and Generation

The rconstruction of realistic 3D human is crucial in VR/AR applications. This talk focuses on 3D human modeling, covering topics from traditional volumetric capture to neural rendering, from per-scene optimization to generalizable prior model training and generative methods.

Dec. 26, 2024, ByteDance, Online Live Stream

Slide WeChat Post

Awards

• PICO "Star Team Award" Innovation Breakthrough Award (创新突破奖 ), Bytedance, 2023
• Tencent Open Source Collaboration Award (腾讯开源协同奖 ), Tencent, 2021
• Outstanding Graduate of Beijing (北京市优秀毕业生 ), Beijing, 2021
• Outstanding Graduate of Department of Automation, Tsinghua University, 2021
• Excellent Bachelor Thesis Award, Northeastern University, 2018
• Outstanding Graduate of Liaoning Province, Liaoning Province, 2018
• National Scholarship, Ministry of Education, 2018
• Excellence Award for National Undergraduate Innovation Program, Northeastern University, 2017
• City's Excellent Undergraduate, Shenyang City, 2017
• Mayor's Scholarship, Shenyang City, 2017
• Top Ten Excellent Undergraduate (10 / the whole university, 十佳本科生 ), Northeastern University, 2017
• Honorable Mention of American Mathematical Contest in Modeling, COMAP, 2017
• Second Prize of National Undergraduate Mathematical Contest in Modeling, CSIAM, 2016
• First Prize of Provincial Undergraduate Mathematical Contest in Modeling, Liaoning Province, 2016
• 2x Second Prize of Electronic Design Contest, Education Department of Liaoning Province, 2015-2016
• 4x First Class Scholarships, Northeastern University, 2015-2018