The 2025 Conference on Computer Vision and Pattern Recognition (CVPR 2025) recently took place in Nashville, Tennessee, the United States. Prof Xu Mingliang's team from the School of Computer Science and Artificial Intelligence at Zhengzhou University (ZZU) clinched the championship at the NTIRE 2025 Cross-Domain Few-Shot Object Detection Challenge workshop, showcasing the university's international competitiveness in cutting-edge AI research.
The challenge focused on the international research conundrum of Cross-Domain Few-Shot Object Detection (CD-FSOD) in computer vision, requiring the transfer of source domain knowledge to a new target domain with minimal annotated samples and significant domain differences.
The team innovatively proposed the CD-FSOD method based on Instance Feature Cache, effectively overcoming knowledge transfer barriers posed by cross-domain disparities. They secured first place in the closed-track competition on three highly challenging test sets, surpassing numerous top international teams.
CVPR is the largest and most influential international academic conference in the field of AI, classified as a category A conference by the China Computer Federation (CCF). Its workshops and challenges focus on the performance of cutting-edge technologies in high-difficulty application evaluations, attracting leading global tech companies, top scholars, and industry experts and serving as a crucial platform for showcasing the latest innovative achievements.
Xu Mingliang's team from ZZU clinches the championship at the CVPR 2025. [Photo/zzu.edu.cn]