I am a third-year Ph.D. student in Artificial Intelligence at Zhejiang University, advised by Prof. Yi Yang. I am also a research intern at ByteDance, working on video generation and editing.
My research centers on generative visual intelligence, with a particular interest in connecting visual generation with geometry. My work spans controllable image and video generation, 3D vision, transparent-object geometry estimation, Gaussian Splatting, and digital human reconstruction and generation.
- Video Generation and Editing — generative modeling, RGBA generation, visual understanding, and controllable content creation
- Controllable Image Generation — multi-condition alignment and fine-grained instance control
- 3D Vision and Gaussian Splatting — transparent-surface reconstruction and efficient neural rendering
- Geometry Estimation — surface normal and depth estimation for challenging real-world materials
- Digital Humans — efficient reconstruction, generation, and real-time rendering
- May 2026 — TransNormal was accepted to ICML 2026.
- Jan 2026 — BideDPO was accepted to ICLR 2026.
- May 2025 — DreamRenderer was accepted to ICCV 2025.
- July 2025 — TSGS was accepted to ACM Multimedia 2025 as an Oral Presentation.
Mingwei Li, Hehe Fan, Yi Yang
ICML 2026
Single-step transparent-object surface normal estimation with dense visual semantics and diffusion priors.
Dewei Zhou, Mingwei Li, Zongxin Yang, Yu Lu, Yunqiu Xu, Zhizhong Wang, Zeyi Huang, Yi Yang
ICLR 2026
A bidirectionally decoupled preference optimization framework for resolving conflicts between text and visual conditions.
Dewei Zhou, Mingwei Li, Zongxin Yang, Yi Yang
ICCV 2025
A training-free controller for precise multi-instance attribute binding in large-scale text-to-image models.
TSGS: Improving Gaussian Splatting for Transparent Surface Reconstruction via Normal and De-lighting Priors
Mingwei Li, Pu Pang, Hehe Fan, Hua Huang, Yi Yang
ACM Multimedia 2025 Oral
A two-stage Gaussian Splatting framework for geometrically accurate and photorealistic transparent-surface reconstruction.
Mingwei Li, Jiachen Tao, Zongxin Yang, Yi Yang
Fast monocular dynamic-human reconstruction with 100+ FPS rendering and training completed in hundreds of seconds.
- National First Prize, Huawei Cup Graduate AI Innovation Competition — Team Leader, 4th of 2,618 teams (Top 0.15%), 2024