MPU Student's Research Selected for CVPR, the World's Top Academic Conference on Computer Vision（Image source: CVPR 2025）

Li Yanfeng, a PhD candidate in the Computer Applied Technology Programme at the Faculty of Applied Sciences of the Macao Polytechnic University (MPU)

Li Yanfeng, a PhD candidate in the Computer Applied Technology programme at the Faculty of Applied Sciences of the Macao Polytechnic University (MPU), collaborated with Dr. Chan Ka Hou, Lecturer Sun Yue, Professor Lam Chan Tong, Associate Professor Tan Tao, as well as researchers from the John Hopcroft Center for Computer Science at Shanghai Jiao Tong University, the Greater Bay University, Fuzhou University, and Sichuan University to conduct joint research and propose an innovative image editing algorithm named MoEdit. This research achievement was successfully accepted for presentation at CVPR (Computer Vision and Pattern Recognition) 2025—the top academic conference in the field of computer vision—fully demonstrating MPU's research strength in artificial intelligence (AI) and image processing.

This research proposes solutions to the core challenges currently faced in multi-object image editing. Traditional image editing methods fail to simultaneously balance the unique attributes of each object and the structural consistency of the overall scene when processing complex scenes containing multiple objects, leading to issues such as quantity errors, visual distortion, and ambiguous object attributes in editing results. In addition, conventional editing methods usually require auxiliary tools such as masks and bounding boxes for operation, which increases operational complexity and limits the flexibility of application scenarios.

To address the above problems, the research team designed two key modules: Feature Compensation (FeCom) and Quantity Attention (QTTN). The FeCom module can effectively separate and preserve the independent features of each object, avoiding feature confusion between objects; the QTTN module automatically perceives and maintains the quantity consistency of objects in images through the image information enhanced by FeCom, without relying on external auxiliary tools. This breakthrough enables MoEdit to perform excellently in multiple image editing tasks such as style transfer, object reshaping, and background reconstruction. The research outcome provides efficient, intelligent, and easy-to-operate image processing tools for content creators, designers, and industries reliant on visual content such as e-commerce, and is expected to be widely applied in fields including digital media production, product display, and intelligent image processing.

CVPR, hosted by the Institute of Electrical and Electronics Engineers (IEEE), is one of the most prestigious top-tier academic conferences in the global field of computer vision and AI, and is rated as a Class A conference by the China Computer Federation (CCF). According to Google Scholar statistics, CVPR currently has an h5-index of 440 and an h5-median of 689, ranking among the top of all academic conferences worldwide. It ranks first globally among computer vision conferences and second only to Nature among all types of publications, making it one of the most influential academic events in the field.