EXPLORING VISUAL INTELLIGENCE

Mingwei Li.

Seeing beyond
the visible.

Bridging generative intelligence and the 3D world.
Ph.D. student at Zhejiang University.

Explore my research Google Scholar
LATENT GEOMETRY001 / ∞
SCROLL TO DISCOVER HANGZHOU, CHINA UTC +08:00
RESEARCH & LEARNINGZhejiang UniversityBeijing Zhongguancun AcademyByteDance

01 / A LITTLE ABOUT ME

Curiosity, in every dimension.

Mingwei Li
Mingwei Li / 李明伟

Ph.D. Student · AI Researcher

I build models that understand
and create the visual world.

I am a third-year Ph.D. student at Zhejiang University in Hangzhou, China, advised by Prof. Yi Yang. My research centers on generative visual intelligence, spanning controllable image and video generation, 3D vision, transparent-object geometry estimation, and digital humans. As part of my doctoral training, I am jointly trained with Beijing Zhongguancun Academy under the mentorship of Prof. Tie-Yan Liu.

I am also a research intern at ByteDance, working on video generation and editing. Prior to that, I obtained my B.Sc. in Artificial Intelligence from Zhejiang University in 2023 (Outstanding Graduate, ranked 2nd in major). I was also a member of the Mixed Class at Chu Kochen Honors College (CKC) of Zhejiang University.

GitHub ↗Google Scholar ↗Email ↗
01

3D / 4D Reconstruction

Multi-view stereo, neural radiance fields, and Gaussian splatting for high-quality scene reconstruction.

GEOMETRY · NeRF · GAUSSIANS
02

Transparent Understanding

Surface normal estimation for transparent and reflective objects using foundation models.

NORMALS · DIFFUSION · SEMANTICS
03

Video Generation & Editing

Generative modeling, RGBA generation, visual understanding, and controllable content creation.

GENERATION · CONTROL · DIGITAL HUMANS

02 / SELECTED PUBLICATIONS

Ideas into impact.

TransNormal research overviewView figure

TransNormalDense Visual Semantics for Diffusion-based Transparent Object Normal Estimation

Mingwei Li, Hehe Fan, Yi Yang✉

ICML 2026

We leverage dense visual semantics to guide diffusion-based single-step normal prediction with wavelet regularization for transparent object normal estimation.

Paper Project Code Dataset
BideDPO research overviewView figure

BideDPOConditional Image Generation with Simultaneous Text and Condition Alignment

Dewei Zhou, Mingwei Li, Zongxin Yang, Yu Lu, Yunqiu Xu, Zhizhong Wang, Zeyi Huang, Yi Yang✉

ICLR 2026

We propose a bidirectionally decoupled DPO method to resolve text-condition conflicts in controllable text-to-image generation, significantly improving both text and condition adherence.

Paper Project Code OpenReview
TSGS research overviewView figure

TSGSImproving 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

We introduce normal and de-lighting diffusion priors to optimize transparent surface reconstruction, and design a sliding-window depth extraction method to improve geometric accuracy and rendering quality for transparent objects.

Paper Project Code
DreamRenderer research overviewView figure

DreamRendererTaming Multi-Instance Attribute Control in Large-Scale Text-to-Image Models

Dewei Zhou, Mingwei Li, Zongxin Yang, Yi Yang✉

ICCV 2025

This work introduces text attribute hard-binding, image attribute hard-binding, and image attribute soft-binding mechanisms to enhance control precision in conditional text-to-image generation, achieving state-of-the-art on COCO-MIG benchmark.

Paper Project Code
Human101 research overviewView figure

Human101Training 100+FPS Human Gaussians in 100s from 1 View

Mingwei Li, Jiachen Tao, Zongxin Yang, Yi Yang✉

arXiv 2023

Combining 3D Gaussian Splatting with human body priors, we achieve fast dynamic human reconstruction from monocular video with rendering speeds exceeding 100+ FPS.

Paper Project Code

03 / LATEST UPDATES

Notes from the lab.

2026.05
Accepted TransNormal accepted to ICML 2026!
2026.02
Released TransNormal paper, code, and dataset released. [arXiv]
2026.01
Accepted BideDPO accepted to ICLR 2026!
2025.07
Accepted TSGS accepted to ACM Multimedia 2025 as Oral Presentation!
2025.05
Accepted DreamRenderer accepted to ICCV 2025!
2024.10
Award Won National First Prize in the 6th "Huawei Cup" Graduate AI Innovation Competition (Rank 4/2618, Top 0.15%).
2023.12
Released Human101 paper and code released on arXiv.
2023.06
Milestone Graduated from Zhejiang University with B.Sc. in Artificial Intelligence (Outstanding Graduate).

04 / THE JOURNEY

Always learning.

2024.06 - Present

Zhejiang University

Ph.D. in Artificial Intelligence, College of Computer Science and Technology

Advisor: Prof. Yi Yang · Direct Ph.D. from Master's program

Joint training: Beijing Zhongguancun Academy · Mentor: Prof. Tie-Yan Liu

2023.09 - 2024.06

Zhejiang University

M.Sc. in Computer Science and Technology, College of Computer Science and Technology

Advisor: Prof. Yi Yang · Transferred to Ph.D. after one year

2019.09 - 2023.06

Zhejiang University

B.Sc. in Artificial Intelligence, College of Computer Science and Technology

Outstanding Graduate · Ranked 2nd in major · Recommended for graduate school

05 / RECOGNITION

Awards & honors.

2024 National First Prize in the 6th "Huawei Cup" Graduate AI Innovation Competition (Team Leader, Rank 4/2618, Top 0.15%)
2024 Outstanding Graduate Student Cadre & Five-Good Graduate Student, College of Computer Science, Zhejiang University
2023 Outstanding Graduate, Zhejiang University
2023 Outstanding Undergraduate Thesis, College of Computer Science, Zhejiang University
2021 First-class Scholarship & Outstanding Student, Zhejiang University (2019-2020, 2020-2021)
2020 Second Prize in Zhejiang Province College Physics Innovation Competition

ACADEMIC SERVICE

Conference reviewer AAAI ICML

LET’S CONNECT

Great ideas start
with a conversation.

mingweili@zju.edu.cn

Hangzhou, China

GitHub ↗Google Scholar ↗