Comparison of Yolov7 and Yolov8 Architectures for Detecting Shirt Collars

The collar of the shirt is one of the primary aspects monitored during online examinations in the postgraduate program at Gunadarma University. Each examinee is required to wear formal, collared attire. Based on these regulations, a study is needed to develop a collar detection method to facilitate the online exam monitoring process. This research involves a comparative analysis between two detection architectures: You Only Look Once (YOLO) version 7 (YOLOv7) and version 8 (YOLOv8), to determine the most effective architecture for detecting shirt collars according to the dataset used in the study. The detection models developed from both architecture versions are implemented in a web-based application and tested to evaluate their accuracy and efficiency. Based on testing, YOLOv7 demonstrated an average accuracy of 95%, outperforming YOLOv8, which had an average accuracy of 75%. However, although YOLOv8's accuracy was lower than YOLOv7's, YOLOv8 excelled in detection speed, with an average time of 2.27 seconds, significantly faster than YOLOv7's average time of 22.42 seconds. Comparing accuracy and speed, YOLOv7 still shows the best performance in this study. However, it remains possible that YOLOv8 could surpass YOLOv7 in the future, especially if YOLOv8's detection accuracy can be significantly improved.