FCOS-Mask: Fully Convolution Neural Network for Face Mask Detection

Since the end of 2019, the coronavirus disease 2019 (COVID-19) has spread globally, posing enormous challenges to the global society. Wearing a mask has been proven to be the easiest and the most effective way to limit the spread of COVID-19, and it has become the rule in many public areas. This has also led to a growing demand for automatic real-time mask detection services to replace manual reminders. However, current research on mask detection still has limitations, and both the accuracy and speed can be further improved. In this paper, we propose FCOS-Mask, a one-stage anchor-free object detection method for face mask detection. We add a bottom-up feature augmentation path to the model’s neck and conduct Mosaic to strengthen the ability to detect objects in an unusual context. Moreover, we adopt CIoU and Soft-NMS to improve the training speed and detection accuracy on occluded faces. FCOS-Mask is tested on the Face Mask Detection dataset and achieves a higher 2.9% mAP compared to baseline with a real-time speed of 20.6 FPS on RTX 2070.