A Comparative Evaluation of YOLOv7, Faster R-CNN and RetinaNet on MS COCO Dataset

This paper presents an end-to-end approach to object detection using the YOLOv7 architecture, evaluated on the MS COCO dataset. MS COCO offers over 200,000 annotated images spanning 80 object categories, serving as a comprehensive benchmark for real-world detection tasks. A pre-trained YOLOv7 model is fine-tuned and assessed alongside two widely used object detection models—Faster R-CNN and RetinaNet. Comparative performance is analyzed using standard evaluation metrics, including mean Average Precision (mAP) at IoU thresholds ranging from 0.50 to 0.95. The results demonstrate that YOLOv7 consistently outperforms the other models across most object categories, offering a superior balance of speed and accuracy. This study confirms YOLOv7’s effectiveness in real-time object detection on MS COCO dataset.