Jie Zhao and Meng Su, Penn State University, USA

Object detection is a pivotal technology in computer vision that detects multi-class objects with their localizations in an image. It can untangle the enigma of complicated scenes in the real world. Two main algorithms for implementing object detection are Single-Shot Detector (SSD) and Faster RCNN, which have unique structures of deep learning neural networks. This study compares two prominent object detection algorithms: SSD and Faster R-CNN, focusing on Intersection over Union (IoU) thresholds and runtime efficiency. Using COCO data sets with the validation of 2017, we evaluate the bounding box localization and recognition accuracy of both algorithms. By analyzing IoU thresholds and time efficiency, our findings offer insights into selecting optimized algorithms for different object detection tasks.

Object Detection, Single Shot Detector, Faster RCNN, Deep Learning, Intersection of Union (IoU), LSVRC (Large Scale Visual Recognition Challenge), COCO Datasets