|
Chaoyi Zhou I am a PhD student in Computer Science at Clemson University advised by Prof. Siyu Huang. Previously, I obtained my M.S. in Computer Science from the University of Southern California, where I was advised by Prof. Yajie Zhao. I received my B.E. in Computer Science and Technology from Nanjing University of Posts and Telecommunications in 2020. My research focuses on 3D and 4D reconstruction, scene representations, and geometric foundation models. I develop methods that recover scene geometry, motion, and semantics from images and videos, with applications in general scene understanding and endoscopic video analysis. |
|
News2026.10: SCOPE-4D is now available on arXiv! 2026.04: Invited talk at the Clemson AI Research Symposium. 2026.03: I will join UII this summer as a Research Intern! 2026.03: FF3R is accepted to CVPR 2026 Findings Track! 2025.09: Bézier Splatting is accepted to NeurIPS 2025! 2025.03: I will join Microsoft this summer as a Research Intern! 2025.01: Latent Radiance Fields is accepted to ICLR 2025! 2024.09: 3DGS-Enhancer is accepted to NeurIPS 2024 as a Spotlight! 2024.01: I will join the Vision and Learning Lab at Clemson University as a PhD student! |
Honors & Awards |
2025.04: Rising Researcher Award from Clemson University
2025.03: ICLR Student Travel Award
Selected Publications* indicates equal contribution |
OursSM4RT |
SCOPE-4D: Endoscopic 4D Geometry Foundation Models Chaoyi Zhou, Zhongpai Gao, Anwesa Choudhuri, Meng Zheng, Benjamin Planche, Run Wang, Terrence Chen, Siyu Huang, Ziyan Wu Preprint, 2026 project page / arXiv SCOPE-4D jointly predicts camera motion, dense geometry, and 3D tissue trajectories from monocular endoscopic video, separating camera motion from tissue deformation. |
|
FF3R: Feedforward Feature 3D Reconstruction from Unconstrained Views
Chaoyi Zhou, Run Wang, Feng Luo, Mert D. Pesé, Zhiwen Fan, Yiqi Zhong, Siyu Huang CVPR Findings, 2026 project page / arXiv FF3R reconstructs geometry and semantic features from unconstrained multi-view images without annotations, supporting novel-view synthesis, open-vocabulary segmentation, and depth estimation. |
|
Latent Radiance Fields with 3D-aware 2D Representations
Chaoyi Zhou*, Xi Liu*, Feng Luo, Siyu Huang ICLR, 2025 project page / arXiv / code Latent Radiance Fields reconstruct 3D scenes directly in a 2D latent space, enabling high-quality novel-view synthesis, including for unbounded scenes. |
|
3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors
Xi Liu*, Chaoyi Zhou*, Siyu Huang NeurIPS, 2024 (Spotlight) project page / arXiv / code 3DGS-Enhancer uses video diffusion priors to restore consistent details across novel views and refine 3D Gaussian Splatting reconstructions. |
Other Publications |
 |
Lang-SVG: Hierarchical Image Vectorization with Language Priors Xi Liu, Chaoyi Zhou, Run Wang, Jiaang Li, Feng Luo, Junxiang Huang, Siyu Huang NeurIPS, 2026 Hierarchical image vectorization with language priors. |
|
Bézier Splatting for Fast and Differentiable Vector Graphics
Xi Liu*, Chaoyi Zhou*, Nanxuan Zhao, Siyu Huang NeurIPS, 2025 project page / arXiv Bézier splatting represents vector graphics with 2D Gaussians sampled along Bézier curves, enabling fast, differentiable rasterization with useful gradients at object boundaries. |
 |
XClipGS: Exact Half-Space Clipping for Medical Volume Gaussian Splatting Zhongpai Gao, Benjamin Planche, Meng Zheng, Anwesa Choudhuri, Chaoyi Zhou, Terrence Chen, Ziyan Wu Preprint, 2026 project page / arXiv / demo XClipGS combines an exact, differentiable clipping operator with interior supervision for Gaussian splatting of medical volumes, revealing internal anatomy with accurate cut surfaces and reduced leakage at interactive rendering speeds. |
 |
Adversarial Prompts for Acceptance Collapse in Speculative Decoding Run Wang, Chaoyi Zhou, Xi Liu, Yi Zhu, Amir Salarpour, Pedram MohajerAnsari, Zhi-Qi Cheng, Feng Luo, Siyu Huang, Mert D. Pesé Preprint, 2026 arXiv ADSD generates adversarial prompt suffixes that reduce token acceptance and slow speculative decoding while preserving task quality, exposing a vulnerability in draft–target model alignment. |
|
FlexMap: Generalized HD Map Construction from Flexible Camera Configurations
Run Wang, Chaoyi Zhou, Amir Salarpour, Xi Liu, Zhi-Qi Cheng, Feng Luo, Mert D. Pesé, Siyu Huang Preprint, 2026 arXiv FlexMap constructs HD maps from flexible camera configurations using a geometry-aware foundation model. It handles missing views and sensor changes without retraining. |
Service |
Conference Reviewer: CVPR, ICCV, ECCV, WACV, ICLR, ICML, NeurIPS, AAAI, Eurographics.