Benchmarking YOLOv3 and SSD: A Performance Comparison for Multi-Object Detection

Abstract

Multiple object detection remains a significant challenge in the field of computer vision. One of the key factors affecting detection performance is the feature extraction process, especially when objects are relatively small or positioned closely together. This study aims to compare the effectiveness of two popular object detection models, YOLO (You Only Look Once) and Single Shot MultiBox Detector (SSD), in detecting multiple objects within images. These models were selected due to their reported high accuracy and real-time processing capabilities, outperforming traditional methods such as the Hough Transform, Deformable Part-based Models (DPM), and conventional CNN architectures. The models were evaluated using a subset of the PASCAL VOC dataset, which includes object categories such as aircraft, faces, cars, and others, with a total of 1,447 annotated images used in training and testing. The evaluation metric used was mean Average Precision (mAP) to assess detection accuracy. Experimental results indicate that YOLO achieves a mAP of 82.01%, while SSD achieves 70.47%. These findings demonstrate that YOLO provides better performance in detecting multiple objects under the same conditions. Overall, this study confirms the advantages of YOLO in scenarios requiring fast and accurate multi-object detection, highlighting its potential for deployment in real-time applications such as autonomous vehicles, surveillance systems, and robotics. The main contribution of this study lies in providing a comparative performance benchmark between YOLO and SSD on a standard multi-object dataset to guide practical model selection in real-time computer vision tasks.

Keywords

• Object Annotation
• Object Detection
• PASCAL VOC
• SSD
• YOLO