Attention-Guided Face Mask Classification Using YOLOv8 with CBAM on MaskedFace-Net Dataset

Abstract

The COVID-19 pandemic has highlighted the critical role of face mask classification in safeguarding public health. While existing computer vision approaches primarily focus on binary mask detection, limited attention has been given to fine-grained multi-class face mask classification, which is more representative of real-world scenarios. In this work, we propose a robust and efficient YOLOv8-CBAM based deep learning framework for classifying three mask-wearing conditions: with mask, without mask, and incorrectly worn mask. Experiments on the MaskedFace-Net dataset demonstrate that the proposed YOLOv8-CBAM model achieves a macro F1-score of 99.71%, with ablation and comparative studies confirming the effectiveness of attention mechanisms, enabling accurate real-time face mask classification for surveillance and edge applications.