Article: "Lite FPN_SSD: A Reconfiguration SSD with Adapting Feature Pyramid Network for Small Object Detection"
Student: Nguyen Dang Duc Manh – 22520847 – KHMT2022.2
Supervisor: Dr. Duong Viet Hang
Summary of the article:
Detecting small objects poses a significant challenge in the field of computer vision due to the low-resolution and blurred features caused by pooling operations across convolutional layers. While single-stage detection algorithms have partially addressed this issue, they still contain redundant features, leading to resource wastage and increased computation time. This research primarily focuses on the SSD algorithm and an enhanced version, Lite FPN-SSD (Lite Single Shot MultiBox Detector with Feature Pyramid Network), proposed to overcome SSD's limitations. The Lite FPN-SSD model utilizes the FPN (Feature Pyramid Network) architecture combined with SSD to create a unified module for controlling features at both shallow and deep levels. With its lightweight design and a slight increase in parameters, the proposed model ensures high efficiency for real-time applications. Extensive experiments were conducted on VOC, VEDAI, and SOHAS datasets, demonstrating impressive results of the proposed models compared to the original SSD and its variants. Notably, with just an additional 2 million parameters, the proposed model achieved an average precision (mAP) of 78.36% on the VOC dataset, approaching another architecture that achieved an mAP of 78.40% but required more than 2.6 million parameters.
I express my deep gratitude to Dr. Duong Viet Hang, my supervisor, for guiding me throughout this research. She has consistently supported and answered my queries during the research process and publication of this international scientific paper. Moreover, she has been a source of motivation, inspiring my passion and dedication to complete this study. I also want to extend my thanks to the Research Fund of the University of Information Technology (UIT-VNUHCM) for the financial support provided for my research project.
The SOICT 2023 Conference (Symposium on Information and Communication Technology) is an annual international conference focusing on new research in critical areas such as AI Foundations and Big Data, Network Communication and Security, Image and Natural Language Processing, Software Engineering and Digital Technology, Blockchain, and Operations Research trends. The conference, technically sponsored by ACM, is the 12th international edition and will take place in Ho Chi Minh City. It is organized by the School of Information and Communication Technology - Hanoi University of Science and Technology, University of Science - Vietnam National University Ho Chi Minh City, Laboratory Informatics, Modelling and Optimisation System (LIMOS), The French National Centre for Scientific Research (CNRS), and Institute of Mathematics - Vietnam Academy of Science and Technology, on December 7-8, 2023. The goal of the SOICT conference is to provide an academic forum for researchers and postgraduate students to share the latest research findings and identify challenges in future computer science. The conference is listed in the reputable Scopus conference list.
For more details, visit: https://www.facebook.com/UIT.Fanpage/posts/pfbid02Vjnp15cby...
Ha Bang - Media Collaborator, University of Information Technology
Translation: Nhat Hien