Small-object detection and its applications

• Main Author: Ho, Yong Xian
• Other Authors: Cuong Dang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Small object detection; YOLO; Deep learning; Faster R-CNN; Autonomous vehicles; Neural networks; Slicing aided hyper inference
• Online Access: https://hdl.handle.net/10356/176733
• Institution: Nanyang Technological University

Small-object detection is a crucial task in computer vision with numerous applications ranging from medical imaging to autonomous driving. This report explores the efficacy of YOLOv8.1, a state-of-the-art single-stage object detector, specifically focusing on its capabilities in detecting small objects in the context of autonomous vehicles. Various techniques were employed to enhance small object detection performance, including Slicing Aided Hyper Inference (SAHI), varying input image size and class weight adjustments. Additionally, traditional two-stage detectors such as Faster R-CNN are also evaluated for comparison. The results reveal that SAHI and Faster R-CNN exhibit excellent capabilities but suffer from slow inference speeds due to computational complexities. Conversely, YOLOv8.1 trained using full image resolution with class weight adjustment is the most effective solution for small object detection, offering the best trade-off between speed and accuracy. This study underscores the importance of tailored optimisation strategies for small-object detection tasks.