ICCV 2025 (Oral)

Self-Ensembling Gaussian Splatting for Few-Shot Novel View Synthesis

Chen Zhao, Xuan Wang, Tong Zhang, Saqib Javed, Mathieu Salzmann

Project |

• We observe that 3D Gaussian Splatting (3DGS) excels in novel view synthesis (NVS) but overfits with sparse views, so we propose Self-Ensembling Gaussian Splatting (SE-GS) using an uncertainty-aware perturbation strategy to train a main model alongside perturbed models. We minimize discrepancies between these models to form a robust ensemble for novel-view generation.

ICCV 2025

Fine-Grained 3D Gaussian Head Avatars Modeling from Static Captures via Joint Reconstruction and Registration

Yuan Sun, Xuan Wang, Cong Wang, WeiLi Zhang, Yanbo Fan, Yu Guo, Fei Wang

• 3D Gaussian-based head avatar modeling performs well with enough data, but prior-based methods lack rendering quality due to limited identity-shared representation power. We solve this via joint reconstruction and registration of prior-based and prior-free 3D Gaussians, merging and post-processing for complete avatars, with experiments showing better quality and high-resolution support.

ICCV 2025

Diffuman4D: 4D Consistent Human View Synthesis from Sparse-View Videos with Spatio-Temporal Diffusion Models

Yudong Jin, Sida Peng, Xuan Wang, Tao Xie, Zhen Xu, Yifan Yang, Yujun Shen, Hujun Bao, Xiaowei Zhou

Project |

• This paper addresses high-fidelity novel-view synthesis from sparse-view human videos. While 4D diffusion models tackle limited observations, they lack spatio-temporal consistency, so we propose a sliding iterative denoising process on a latent grid (encoding image, camera and human poses) via alternating spatial-temporal denoising, enabling sufficient information flow and affordable GPU memory.

ICCV 2025

DGTalker: Disentangled Generative Latent Space Learning for Audio-Driven Gaussian Talking Heads

Xiaoxi Liang, Yanbo Fan, Qiya Yang, Xuan Wang, Wei Gao, Ge Li

• In this work, we investigate generating high-fidelity, audio-driven 3D Gaussian talking heads from monocular videos, presenting DGTalker—a real-time, high-fidelity, 3D-aware framework that uses Gaussian generative priors and latent space navigation to alleviate 3D information lack and overfitting issues. We propose a disentangled latent space navigation framework for precise lip and expression control, plus masked cross-view supervision for robust learning. Extensive experiments show DGTalker outperforms state-of-the-art methods in visual quality, motion accuracy, and controllability.

CVPR 2025

AvatarArtist: Open-Domain 4D Avatarization

Hongyu Liu, Xuan Wang$^\dagger$, Ziyu Wan, Yue Ma, Jingye Chen, Yanbo Fan, Yujun Shen, Yibing Song, Qifeng Chen$^\dagger$

Project |

• This work focuses on open-domain 4D avatarization, with the purpose of creating a 4D avatar from a portrait image in an arbitrary style. Extensive experiments suggest that our model, termed AvatarArtist, is capable of producing high-quality 4D avatars with strong robustness to various source image domains.

CVPR 2025

HERA: Hybrid Explicit Representation for Ultra-Realistic Head Avatars

Hongrui Cai$^\star$, Yuting Xiao$^\star$, Xuan Wang$^\dagger$, Jiafei Li, Yudong Guo, Yanbo Fan, Shenghua Gao, Juyong Zhang$^\dagger$

• We present a hybrid explicit representation to combine the strengths of different geometric primitives, which adaptively models rich texture on smooth surfaces as well as complex geometric structures simultaneously.
• To avoid artifacts created by facet-crossing Gaussian splats, we design a stable depth sorting strategy based on the rasterization results of the mesh and 3DGS.
• We incorporate the proposed hybrid explicit representation into modeling 3D head avatars, which render more fidelity images in real time.

ECCV 2024

Real-time 3d-aware portrait editing from a single image

Qingyan Bai, Zifan Shi, Yinghao Xu, Hao Ouyang, Qiuyu Wang, Ceyuan Yang, Xuan Wang, Gordon Wetzstein, Yujun Shen, Qifeng Chen

Project

• This work presents 3DPE, a practical method that can efficiently edit a face image following given prompts, like reference images or text descriptions, in a 3D-aware manner.

SIGGRAPH 2024 (Conf)

HeadArtist: Text-conditioned 3d head generation with self score distillation

Hongyu Liu, Xuan Wang$^\dagger$, Ziyu Wan, Yujun Shen, Yibing Song, Jing Liao, Qifeng Chen$^\dagger$

Project |

• We present HeadArtist for 3D head generation following human-language descriptions. With a landmark-guided ControlNet serving as a generative prior, we come up with an efficient pipeline that optimizes a parameterized 3D head model under the supervision of the prior distillation itself. We call such a process self score distillation (SSD).

ICCV 2023

ToonTalker: Cross-Domain Face Reenactment

Yuan Gong, Yong Zhang, Xiaodong Cun, Fei Yin, Yanbo Fan, Xuan Wang, Baoyuan Wu, Yujiu Yang

• We propose a novel method for cross-domain reenactment without paired data.

SIGGRAPH 2023 (Conf)

NOFA: NeRF-based One-shot Facial Avatar Reconstruction

Wangbo Yu, Yanbo Fan$^\dagger$, Yong Zhang$^\dagger$, Xuan Wang$^\dagger$, Fei Yin, Yunpeng Bai, Yan-Pei Cao, Ying Shan, Yang Wu, Zhongqian Sun, Baoyuan Wu

• We propose a one-shot 3D facial avatar reconstruction framework, which only requires a single source image to reconstruct high-fidelity 3D facial avatar, by leveraging the rich generative prior of 3D GAN and developing an efficient encoder-decoder network.

CVPR 2023 (Highlight)

Next3D: Generative Neural Texture Rasterization for 3D-Aware Head Avatars

Jingxiang Sun, Xuan Wang, Lizhen Wang, Xiaoyu Li, Yong Zhang, Hongwen Zhang, Yebin Liu

Project |

• We propose a 3D representation called Generative Texture-Rasterized Tri-planes that learns Generative Neural Textures on top of parametric mesh templates and then projects them into three orthogonal-viewed feature planes through rasterization, forming a tri-plane feature representation for volume rendering.

CVPR 2023

High-fidelity Facial Avatar Reconstruction from Monocular Video with Generative Priors

Yunpeng Bai, Yanbo Fan, Xuan Wang, Yong Zhang, Jingxiang Sun, Chun Yuan, Ying Shan

• We propose a new method for NeRF-based facial avatar reconstruction that utilizes 3D-aware generative prior. Different from existing works that depend on a conditional deformation field for dynamic modeling, we propose to learn a personalized generative prior, which is formulated as a local and low dimensional subspace in the latent space of 3D-GAN.

CVPR 2023

3D GAN Inversion with Facial Symmetry Prior

Fei Yin, Yong Zhang, Xuan Wang, Tengfei Wang, Xiaoyu Li, Yuan Gong, Yanbo Fan, Xiaodong Cun, Ying Shan, Cengiz Oztireli, Yujiu Yang

Project |

• We propose a novel method to promote 3D GAN inversion by introducing facial symmetry prior.

CVPR 2023

High-Fidelity Clothed Avatar Reconstruction from a Single Image

Tingting Liao, Xiaomei Zhang, Yuliang Xiu, Hongwei Yi, Xudong Liu, Guo-Jun Qi, Yong Zhang, Xuan Wang, Xiangyu Zhu, Zhen Lei

Project

• By combining the advantages of the high accuracy of optimization-based methods and the efficiency of learning-based methods, we propose a coarse-tofine way to realize a high-fidelity clothed avatar reconstruction (CAR) from a single image.

SIGGRAPH Asia 2022 (Conf)

VideoReTalking: Audio-based Lip Synchronization for Talking Head Video Editing In the Wild

Kun Cheng, Xiaodong Cun, Yong Zhang, Menghan Xia, Fei Yin, Mingrui Zhu, Xuan Wang, Jue Wang, Nannan Wang

Project | 🔥

• VideoReTalking, a new system to edit the faces of a real-world talking head video according to an input audio, producing a high-quality and lip-syncing output video even with a different emotion.

ECCV 2022

StyleHEAT: One-Shot High-Resolution Editable Talking Face Generation via Pre-trained StyleGAN

Fei Yin, Yong Zhang, Xiaodong Cun, Mingdeng Cao, Yanbo Fan, Xuan Wang, Qingyan Bai, Baoyuan Wu, Jue Wang, Yujiu Yang

Project |

• We propose a novel unified framework based on a pre-trained StyleGAN that enables a set of powerful functionalities, i.e., high-resolution video generation, disentangled control by driving video or audio, and flexible face editing.

CVPR 2022

HDR-NeRF: High Dynamic Range Neural Radiance Fields

Xin Huang, Qi Zhang, Ying Feng, Hongdong Li, Xuan Wang, Qing Wang

Project |

• High Dynamic Range Neural Radiance Fields (HDR-NeRF) to recover an HDR radiance field from a set of low dynamic range (LDR) views with different exposures.