Yuhai Deng 邓宇海

I am currently pursuing an M.S. degree at the College of Computer Science, Nankai University, under the supervision of Prof. Xiang Li, whose insightful guidance has greatly shaped my research interests and approach.
Prior to that, I received my B.E. degree from Central South University in 2024, where I was honored as an Outstanding Graduate. During my undergraduate studies, I had the privilege of working with Prof. Shichao Kan, who taught me how to approach research problems and introduced me to the world of computer vision.

I believe intelligence is inherently multimodal. Models should be able not only to understand the world, but also to think in vision. Intelligent models should unify visual representation, visual reasoning, and visual generation, bridging understanding, thinking, and creation.

yuhai_deng@126.com GitHub

Research Experience

My research experience spans visual retrieval, reference-guided image tone editing, and expressive talking-face generation. Across these projects, I have primarily explored diffusion-based generation and representation learning.

Reference-Guided Image Tone Editing

Developed ICTone to explore semantic-aware image tone editing from a reference image through diffusion-based generation.

Object Retrieval

Studied object retrieval for outside-knowledge VQA and developed MS-GCEL for long-tailed visual retrieval.

Expressive Talking-Face Generation

Developed VAExpress to explore expressive talking-face generation with audio-aligned facial expressions.

ECCV 2026 First Author

Towards In-Context Tone Style Transfer with A Large-Scale Triplet Dataset

Yuhai Deng, Huimin She, Wei Shen, Meng Li, Ruoxi Wu, Lunxi Yuan, Xiang Li

A diffusion framework for reference-guided image tone editing that transfers semantically corresponding tones from a reference image through in-context generation.

Project Page arXiv Code Dataset

Figure 8: VQA examples comparing GPT-4o without retrieval, with STML, and with MS-GCEL

arXiv Preprint Student First Author

Object Retrieval for Visual Question Answering with Outside Knowledge

Shichao Kan, Yuhai Deng, Jiale Fu, Lihui Cen, Zhe Qu, Linna Zhang, Yixiong Liang, Yigang Cen

Introduces object retrieval for outside-knowledge VQA and MS-GCEL, an unsupervised multi-scale collaborative embedding method for robust retrieval of long-tailed objects.

arXiv Code HTML

VAExpress results showing audio-aligned facial expressions in expressive talking-face generation

Practice Project

VAExpress: Expressive Talking-Face Generation with Continuous Emotion Control

Yuhai Deng

A practical study on expressive talking-face generation that transforms a single portrait and an audio clip into a natural video with audio-aligned facial expressions.

Project Page Technical Report

Education

M.S. Student in Computer Science and Technology | Nankai University

Time: 2024.09 — Present

B.E. in Intelligent Science and Technology | Central South University

Time: 2020.09 — 2024.06

Honors & Awards

Jun. 2024

Outstanding Graduate

Central South University

Aug. 2023

Second Prize

The 16th Chinese Collegiate Computing Competition

Dec. 2021

National Scholarship

Central South University