Image Restoration & Enhancement
Designing general and efficient models that recover faithful, high-quality images from real-world degradation, from lightweight super-resolution to large-scale restoration.
Research portfolio · 2026
Zheyuan LI
Advancing image restoration, visual quality understanding, and controllable generative models.
Currently a Ph.D. student at University of Macau, supervised by Prof. Jiantao Zhou.
Also a research assistant in the XPixel Group, working with Prof. Chao Dong and Dr. Xiangyu Chen.
Chapter I · Research
Research Interests
My work develops capable and efficient visual systems, with an emphasis on understanding how images are restored, evaluated, and generated.
Vision-Language Quality Assessment
Moving beyond scalar scores toward descriptive, comparative, and human-aligned reasoning about visual quality with multimodal language models.
Controllable Generative Models
Building efficient control mechanisms for large diffusion models while studying the information flow that makes precise visual generation possible.
Chapter II · Selected Work
Selected Work
A selection of work spanning image quality assessment, restoration, model interpretation, and controllable generation.
Enhancing Descriptive Image Quality Assessment With a Large-Scale Multi-Modal Dataset
A multi-functional VLM-based IQA model with a 495K large-scale dataset, surpassing GPT-4V
Towards Efficient SDRTV-to-HDRTV by Learning from Image Formation
An efficient three-step SDRTV-to-HDRTV conversion method with only 35K parameters for global color mapping
UniCon: Unidirectional Information Flow for Effective Control of Large-Scale Diffusion Models
Unidirectional information flow reduces GPU memory by 1/3 and speeds up training by 2.3x for diffusion model control
Interpreting Low-Level Vision Models With Causal Effect Maps
Introduce causality theory to interpret low-level vision models with a model-/task-agnostic method
A Comparative Study of Image Restoration Networks for General Backbone Network Design
Propose a general backbone for image restoration
Depicting Beyond Scores: Advancing Image Quality Assessment through Multi-modal Language Models
Propose a language-based image quality assessment method
Chapter III · Biography
Personal Experience
My path combines academic training, collaborative research, and practical work on efficient visual intelligence.
Since 2022, I have worked as a research assistant in the XPixel Group at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences.
I received my bachelor's degree in Software Engineering from Northwestern Polytechnical University and collaborate closely with researchers working across image restoration, quality assessment, and generative modeling.
- Ph.D. in Computer ScienceUniversity of Macau
- Research AssistantXPixel Group · SIAT, Chinese Academy of Sciences
- B.S. in Software EngineeringNorthwestern Polytechnical University
Chapter IV · Updates
News & recognition.
-
New work on descriptive image quality assessment, causal interpretation, and diffusion model control appeared in TIP, TPAMI, and ICLR.
-
Started Ph.D. study at the University of Macau.
-
Won the NTIRE Efficient Super-Resolution Challenge model complexity track with BSRN-S.