Comparative Analysis of YOLOv4 and YOLOv4-tiny Techniques towards Face Mask Detection
Ravi Anand, Jasowanta Das, Pratima Sarkar · 2021 International Conference on Computational Performance Evaluation (ComPE) · 2021
Covid-19 has brought various complications in our day-to-day life leading to a disruption in overall movements across the world. Although still researchers and scientists are working on finding more effective ways to deal with it, wearing a face is one of the most simplistic yet efficient ways to overcome this. Wearing a face mask all the time in public places has become a new normal. Therefore, face mask detection for monitoring of people in public places has become a crucial task. Deep learning has been used to make recent advances in the field of object detection. To accomplish this objective, this research employs three state-of-the-art object identification models, notably YOLOv4 and YOLOv4-tiny. The models were trained using a dataset that included photos of persons wearing and not wearing masks. Considering it for surveillance purposes, it can also be used for detection of face and mask in motion. The models employ an approach that involves drawing bounding boxes (red or green) around people’s faces and determining whether or not they are wearing a face mask. Further, the performance of these models was compared using mAP, recall F1-score and FPS