CityU FDS holds exhibition of outstanding undergraduate graduation projects, students showcase remarkable sci-tech innovations
The Selection & Exhibition of Outstanding Undergraduate Final-Year Project 2025/2026 of the Faculty of Data Science (FDS) at City University of Macau (CityU) was held on Taipa Campus. The selected works focus on leading research areas including artificial intelligence (AI), medical image recognition, and large language model (LLM), demonstrating the innovations of the undergraduate students in their graduation projects. The exhibition fully showcased the remarkable achievement of CityU in cultivating undergraduate talent.
A total of 20 outstanding graduation projects were selected after a rigorous selection process. During the exhibition, a professional panel further selected 11 of the most representative works to receive the “Outstanding Graduation Project Award”. The exhibited works cater to industry hotspots and social needs, and covered multiple dimensions including technological development, real-world application, algorithm optimization, and system construction, reflecting the students’ professional competence and innovative ability in integrating theory with practice.
During the visit, the panel and faculty members learned about the creative concepts, technological breakthroughs, and practical application scenarios of the projects. They expressed appreciation for the students’ innovative thinking in interdisciplinary fields, cutting-edge algorithm optimization, and efforts to address social problems and issues, and engaged in in-depth exchanges with the students regarding technology application and future directions for optimization.
The graduation projects exhibited this year feature diverse topics and broad coverage, following the development trends in data science and AI industries, and covering innovative practices across numerous fields including super-resolution reconstruction of meteorological element, localized deployment of lightweight LLM for financial stock selection, multi-task intelligent analysis of diabetic retinopathy, real-time voice cloning and multilingual synthesis, VAR-based generative image editing. A considerable number of works have achieved innovative breakthroughs in model accuracy, low cost application, privacy security, and adaptation to real-world scenarios, demonstrating the solid professional foundation, research exploration capabilities, and cross-disciplinary engineering innovation strength of the undergraduate students. These also evidenced the distinctive approach to cultivating high-quality, practice-oriented, and innovative talents.
Selected Projects
Diffusion-Based Fine-Grained Attribute Slider Control
Developer: Zhou Zirou | Supervisor: Assistant Prof. Zhu Hongguang
This project develops a controllable image generation system using Stable Diffusion 1.5 and LoRA, featuring an innovative dual‑LoRA fusion method for precise, continuous control of visual attributes. The system is lightweight, compatible with standard devices, and easy to use — effectively addressing the challenge of fine‑grained image generation with strong practical application value.
VAR-Based Generative Image Editing
Developer: Cheng Yifan | Supervisor: Assistant Prof. Zhu Hongguang
Project Description:
This project reproduces and deploys the lightweight VAREdit-8B-512 visual autoregressive model. Using prompt engineering, it extends the model to support diverse artistic styles — including ink wash, oil painting, watercolour, and anime — achieving efficient style transfer while preserving image structure. An interactive editing platform built with Gradio reduces computational costs and enhances the model's practical applicability.
Fundus Image Quality Enhancement Using a Residual Conditional Generative Adversarial Network
Developer: Huang Yuting | Supervisor: Assistant Prof. Liu Chi
Project Description:
This project develops a fundus image quality enhancement method using a residual conditional GAN. To address clinical image degradation (blur, noise), a 10-class random degradation simulation and two-stage training framework were designed. With ROI masks and colour enhancement priors, the method outperforms traditional approaches in structural similarity and perceptual quality — significantly improving image usability for diagnostic tasks.
Spatio‑Temporal Data‑Driven Taxi Cruising Recommendation System
Developer: Cai Yuxing | Supervisor: Assistant Prof. Li Xingfeng
Project Description:
This project enhances taxi cruising efficiency by building a recommendation system and heatmap visualisation platform based on historical order data, using H3 grids, time windows, and a three‑strategy scoring model — providing data‑driven support for driver decisions and traffic management.
Selection & Exhibition of Outstanding Undergraduate Final-Year Project 2025/2026
Website
Scan to learn more!
Source: Faculty of Data Science
