Artificially simulated masked face dataset

Published: 16 March 2021| Version 1 | DOI: 10.17632/nm95xp38rf.1
Contributor:
Shao Liu

Description

This dataset contains faces and masked faces detected and cropped by YOLOv3 from existing public datasets and videos. The cropped faces are in various size and resolutions and they are labeled as no-mask face and masked face. It contains 835 masked faces and 1227 no-mask faces with large variety of image sizes. The largest size is 608x883 and the smallest is 3x4.

Files

Steps to reproduce

1) fine-tuned the YOLOv3 model by WIDERFACE. This dataset contains 32,203 images and 393,703 labeled faces with a high degree of variability in scale, pose and occlusion as depicted in the sample images. 2) The fine-tuned YOLOv3 model can be used to detect and crop faces on images and videos with masked and no-masked faces.