Black Face Mask Dataset

Published: 10 September 2021| Version 2 | DOI: 10.17632/s6rzjp5zcp.2


We present a real Covid-19 face mask dataset of local black-coloured people in Nigeria. The dataset consists of 2,873 images in jpg format, with 3,855 annotations of faces into eight distinct classes and thirteen different annotation file formats. The participants are first grouped by gender (male and female) and then grouped into four different face mask-wearing positions of "No_FaceMask", "Proper_FaceMask", "Improper_FaceMask_1", and "Improper_FaceMask_2". In both male and female "No_FaceMask" classes, the participants did not put on any form of face mask to prevent the spread of the Covid-19 virus; in the case of "Proper_FaceMask", these participants are wearing face masks properly with both nose and mouth adequately covered. For the class "Improper_FaceMask_1" of both males and females, the participants wore the face mask with their noses exposed. Finally, in the last two classes of "Improper_FaceMask_2" the participants incorrectly wore face masks while exposing their nose and mouth. The images are all sizes 640 x 640 x 3, with an average of 0.41 megapixels. For class balance analysis, the number of faces per class is as follows; "Male_No_FaceMask" = 700, "Male_Proper_FaceMask" = 522, "Male_Improper_FaceMask_1" = 489, "Male_Improper_FaceMask_2" = 494, "Female_No_FaceMask" = 503, "Female_Proper_FaceMask" = 404, "Female_Improper_FaceMask_1" = 362, and "Female_Improper_FaceMask_2" = 377. Data analysis, including image labelling, was done with the aid of Roboflow’s platform ( The dataset was collected from the following locations, Covenant University Ota, Ogun State, Taraba State University, Jalingo, Taraba State, and Delta State University, Abraka, Delta State, all in Nigeria. These universities represent collections of students from different parts of the country. The dataset is available in the following thirteen computer vision annotation formats; COCO, CreateML, YOLO Darknet, YOLOv3 Keras, YOLOv4 PyTorch, YOLOv5 PyTorch, Tensorflow Object Detection, RetinaNet Keras, OpenAI Clip Classification, and Tensorflow TFRecords, This dataset aims to complement existing face mask detection datasets by giving a better representation of black-coloured people. Academic researchers can also apply the dataset for occlusion removal and facial recognition research. This is the first face mask dataset that incorporates gender classification with face mask detection in the post-Covid-19 era to the best of our knowledge. The dataset can be merged into smaller classes in applications that do not require such granular classification. The black face mask dataset was generated under the approval and supervision of the Covenant Health Research Ethics Committee (CHREC) and is intended for academic research purpose only.


Steps to reproduce

1. Application for research protocol approval from CHREC. 2. We did image capture with FujiFilm Digital Camera model FInePix S4300 and different smartphones. 3. We did image cropping with PhotoPad Image Editor version 6.43 distributed by NCH Software. 4. Bulk image resizing to 640 x 640 was down with the aid of Pixillion Image Converter, Pixillion Plus version 8.19, distributed by NCH Software. 5. We did image labelling and annotations with the aid of labelImg and Roboflow's label assist. 6. We did dataset Train/Valid/Test split in the ratio of 70:20:10 with the aid of Roboflow's computer vision platform. 7. We did data health analysis with the aid of Roboflow