About
I am a master's student in Computer Science and Technology at Sun Yat-sen University, supervised by Professor Guanbin Li. My research interests center on diffusion-based image and video generation, with an emphasis on video restoration and translation.
During my undergraduate studies, I worked with Associate Professor Xudong Mao, which led to a publication at AAAI 2025 and a granted invention patent.
Education
Sun Yat-sen University
Sep. 2026 – presentM.S. in Computer Science and Technology
Sun Yat-sen University
Sep. 2022 – Jun. 2026B.S. in Artificial Intelligence
News
Released our new preprint RealWeather: Realistic and Scene-Faithful Weather Translation with Driving World Models on arXiv.
Released our new preprint RealityBridge: Bridging Editable 3D Gaussian Splatting Driving Simulations and Real-World Videos on arXiv.
Our paper, CoRe: Context-Regularized Text Embedding Learning for Text-to-Image Personalization, co-authored as the second co-first author, has been accepted at AAAI 2025! project page
Our patent, A Personalized Image Generation Method Based on Text Inversion for Background Embedding and Feature Separation, has been officially granted!
Selected Publications
RealWeather: Realistic and Scene-Faithful Weather Translation with Driving World Models
Yuwei Ning, Liangzhi Wang, Yi Xiao, Zhenhua Wu, Yun Pang, Mingkun Chang, Jichang Li, Guanbin Li†
arXiv preprint · 2026 · Under review
RealWeather is a driving world model for realistic and scene-faithful bidirectional weather translation. It learns authentic weather dynamics directly from real-world videos via Progressive Realism Bootstrapping, and enforces structural integrity with Scene-Fidelity RL Optimization.
RealityBridge: Bridging Editable 3D Gaussian Splatting Driving Simulations and Real-World Videos
Zhenhua Wu, Yun Pang, Mingkun Chang, Yuwei Ning, Liangzhi Wang, Yi Xiao, Guanbin Li†
arXiv preprint · 2026 · Under review
RealityBridge is a structure-preserving and asset-aware Sim-to-Real framework for edited 3DGS driving videos. It uses multimodal controls with a lightweight GateNet for adaptive condition allocation, combined with autoregressive training and reward-guided post-training.
CoRe: Context-Regularized Text Embedding Learning for Text-to-Image Personalization
Feize Wu, Yun Pang, Junyi Zhang, Lianyu Pang, Jian Yin, Baoquan Zhao, Qing Li, Xudong Mao†
AAAI 2025 · 2025
Context Regularization (CoRe) is introduced, which enhances the learning of the new concept's text embedding by regularizing its context tokens in the prompt, thus improving the generalization of the learned text embedding.
Underlined authors contributed equally. † Corresponding author.
Selected Projects
Obsidian plugin
Turn Markdown notes into editable databases in Obsidian, with eight views, formulas, and relations. Everything stays in your local vault.
macOS app
A menu bar app for managing long-running scripts, with live logs, process monitoring, and MCP integration for AI assistants.
Robotics · Course project
A CoppeliaSim simulation platform for comparing path planning and tracking controllers, with differential-drive and Ackermann vehicles and automated experiment reports.