Performance Comparison of Optimizers for YOLOv8n Based Smoker Object Detection

This paper introduces automatic labeling technology to overcome the time-consuming and high cost problems of the existing manual labeling process, with the aim to distinguish between smokers and non-smokers. To address this problem, we present an efficient data labelling method that utilizes computer vision to improve an object detection system for identifying smokers. We also present a comparative analysis of various optimizers - including SGD, Adam, AdamW, Nadam, and RAdam - focusing on their impact on the performance of the YOLOv8n model, which is used to detect smoking behavior. The comparative analysis reveals that the Adam optimizer achieves higher accuracy in detecting objects related to smoking behavior compared to others.