Evaluation of YOLO-X and MobileNetV2 as Face Mask Detection Algorithms

Even though COVID-19 still exists, people are more reluctant to wear masks in public places, in fact only 73% of Indonesian still do. Hence, automatic mask surveillance in public places is still needed. In this paper, we compare two algorithms named YOLO-X and MobileNetV2 to detect face masks. YOLO-X was able to outperform other YOLO algorithms in object detection. While, according to researchers, MobileNetV2 achieved 9S% in face mask detection. To fairly evaluate both algorithms we need to conduct research under controlled variables including using the same datasets and devices. We used public datasets which consists of 1493 mask images and 6451 non mask images for training and testing. The results show that YOLO-X outperforms MobileNetV2 as it achieves 95.0%, 98.7%, 93.7%, and 96.1% for accuracy, average precision, recall, and F1-score respectively. YOLO-X also performs better in detecting faces with occlusion such as glasses, hands, and postures than MobileNetV2. However, YOLO-X detects faces and face masks 31.9% slower than MobileNetV2.