ИССЛЕДОВАНИЕ ПРОИЗВОДИТЕЛЬНОСТИ И СРАВНИТЕЛЬНЫЙ АНАЛИЗ YOLO-NAS И ПРЕДЫДУЩИХ ВЕРСИЙ YOLO

Abstract

This study is devoted to analyzing the performance of the YOLO-NAS (You Only Look Once - Neural Architecture Search) object detection algorithm in comparison with its predecessors from the YOLO family. The aim of the work is to evaluate the performance of YOLO-NAS and to identify its advantages and disadvantages compared to previous versions of the YOLO algorithm. The study is conducted in several key performance aspects, including image processing speed, object detection accuracy, and computational resource utilization efficiency. To achieve these goals, we utilize standard datasets for training and testing object detection algorithms. The research methodology includes the development of an experimental platform that allows us to benchmark the performance of YOLO-NAS and previous versions of YOLO under identical conditions. Experiments are conducted on different hardware and software configurations to evaluate the adaptability of the algorithms to different operating conditions. The results of the study allow us to draw conclusions about the advantages and disadvantages of YOLO-NAS compared to previous versions of YOLO in the context of performance. These conclusions can be useful for computer vision developers and researchers when choosing the most appropriate algorithm for their tasks.