We recently generated a dataset of human faces with correctly and incorrectly worn masks. The objectives are 1) to detect masked faces 2) to classify detected masked faces into correctly masked and incorrectly masked in order to sensitize people about good practices in mask wearing (e.g. in regulated areas monitored by cameras) towards limiting the spread of the COVID-19. This dataset is a base that could also be exploited to generate crowd statistics (e.g. counting people correctly/incorrectly wearing their mask) and other apps for studies in epidemiology, health education and public health.