TransVLM: A Vision-Language Framework and Benchmark for Detecting Any Shot Transitions
Ce Chen, Yi Ren, Yuanming Li, Viktor Goriachko, Zhenhui Ye, Zujin Guo, Zhibin Hong, Mingming Gong

Project | arXiv | GitHub | Blog

• Proposes the Shot Transition Detection (STD) task and TransVLM, a VLM framework that injects optical flow as motion prior for detecting continuous temporal segments of video transitions.
• Deployed to production at HeyGen.

TAVR: Generate Your Talking Avatar from Video Reference
Zujin Guo, Zhenhui Ye, Yi Ren, Yuanming Li, Ce Chen, Zhibin Hong, Chen Change Loy

arXiv

• A novel framework for cross-scene talking avatar generation from video references, integrating token selection and a three-stage training scheme.
• Deployed to production at HeyGen.

Avatar V: Scaling Video-Reference Avatar Video Generation
Benjamin Liang, Ce Chen, Desmond Lin, Ivan Somov, Jiajun Zhao, Jiewei Yuan, Jingfeng Zhang, Junhao Huang, Nik Nolte, Pedram Haqiqi, Penghan Wang, Rong Yan, Rui Zhang, Sam Prokopchuk, Sivan Wang, Viktor Goriachko, Yi Ren, Yuanming Li, Yutao Chen, Zhenhui Ye, Zhibin Hong, Zilong Nie, Zujin Guo

arXiv | Blog

• A production-scale framework for video-reference-conditioned avatar generation. Introduces Sparse Reference Attention for linear-complexity conditioning on long references.
• Deployed to production at HeyGen, trained across thousands of GPUs with a data engine curating 100M+ training clips.

Mobile-VTON: High-Fidelity On-Device Virtual Try-On
Zhenchen Wan*, Ce Chen*, Runqi Lin, Jiaxin Huang, Tianxi Chen, Yanwu Xu, Tongliang Liu, Mingming Gong

Project | arXiv | GitHub |

• A privacy-preserving virtual try-on framework enabling fully offline execution on commodity mobile devices. Introduces a modular TGT architecture with Feature-Guided Adversarial Distillation.
• Matches or outperforms server-based baselines at 1024×768 resolution on VITON-HD and DressCode.

Condition Matters in Full-Head 3D GANs
Heyuan Li, Huimin Zhang, Yuda Qiu, Zhengwentai Sun, Keru Zheng, Lingteng Qiu, Peihao Li, Qi Zuo, Ce Chen, Yujian Zheng, Yuming Gu, Zilong Dong, Xiaoguang Han

Project | arXiv | GitHub

• Proposes view-invariant semantic feature conditioning to decouple 3D head generation from viewing direction, achieving higher fidelity and diversity in full-head synthesis.

CT4D: Consistent Text-to-4D Generation with Animatable Meshes
Ce Chen, Shaoli Huang, Xuelin Chen, Guangyi Chen, Xiaoguang Han, Kun Zhang, Mingming Gong

arXiv

• Presents a novel text-to-4D framework operating on animatable meshes. Introduces the Generate-Refine-Animate (GRA) algorithm for text-aligned mesh generation and uniform driving functions for surface continuity.

SphereHead: Stable 3D Full-head Synthesis with Spherical Tri-plane Representation
Heyuan Li, Ce Chen, Tianhao Shi, Yuda Qiu, Sizhe An, Guanying Chen, Xiaoguang Han

Project | arXiv | GitHub

• Proposes a spherical tri-plane representation in the spherical coordinate system that mitigates “mirroring” artifacts in full-head 3D GAN synthesis, along with a view-image consistency loss.

SCoDA: Domain Adaptive Shape Completion for Real Scans
Yushuang Wu, Zizheng Yan, Ce Chen, Lai Wei, Xiao Li, Guanbin Li, Yihao Li, Shuguang Cui, Xiaoguang Han

Project | arXiv | GitHub

• Proposes the novel SCoDA task for domain adaptive shape completion on real scans, along with the ScanSalon dataset. Introduces cross-domain feature fusion and volume-consistent self-training.

DenseMP: Unsupervised Dense Pre-training for Few-shot Medical Image Segmentation
Zhaoxin Fan, Puquan Pan, Zeren Zhang, Ce Chen, Tianyang Wang, Siyang Zheng, Min Xu

arXiv

• Introduces a two-stage unsupervised dense pre-training pipeline with segmentation-aware dense contrastive learning and superpixel guided pre-training, achieving state-of-the-art few-shot segmentation on Abd-CT and Abd-MRI.

Structure-Preserving Super Resolution with Gradient Guidance
Cheng Ma, Yongming Rao, Yean Cheng, Ce Chen, Jiwen Lu, Jie Zhou

arXiv | GitHub

• Proposes a structure-preserving super resolution method that exploits gradient maps to guide image recovery. Introduces a gradient branch and gradient loss for second-order structural constraints.
• Cited 380+ times.