Article: "Object Detection in Aerial Images Using the Dino Method"
Students involved:
- Le Huu Do - KTPM2022.1 - Lead author
- Ngo Huong Giang - KTPM2022.1 - Co-author
Supervisors:
- MSc. Vo Duy Nguyen
- MSc. Tran Thi Hong Yen
- MSc. Nguyen Thanh Hiep
Summary of the paper:
Recently, data collection from unmanned aerial vehicles has significantly contributed to the development of tasks related to processing aerial images. Three basic tasks in this field include classification, object detection, and segmentation. However, particularly in the task of object detection, researchers often focus on improving traditional models, i.e., models belonging to either one-stage or two-stage categories. However, we have also witnessed the emergence and remarkable development of end-to-end object detection models. Among these methods, DINO (DETR with Improved deNoising anchor boxes) stands out as the first end-to-end method to achieve SOTA (State of the Art) results on the leaderboard. We conducted experiments on the VisDrone dataset with various backbones to comprehensively evaluate the performance of the DINO method and obtained many significant results. This is a promising competitive result for other end-to-end methods.
"We would like to sincerely thank the School of Software Engineering, Multimedia Communications Department, and the UIT-Together research group for providing us with the opportunity to research and complete this paper."
The 26th National Conference on Electronics, Communications, and Information Technology (REV-ECIT) 2023. Date of organization: December 16, 2023.
For more information, please visit: Link
Đông Xanh - Collaborative Communication Officer, University of Information Technology
