Face Mask Detection Video Dataset

Face Mask detection using machine intelligence is one of the hot topic during the COVID-19 period. Variety of works are introduced to identify the policy violators in busy places however, the available image datasets are limited to build a generalized model that can be used in real-time applications with coarse-to-fine video frames.

We present a real-time video/images dataset containing multiple subjects (with/without mask) walking within a University environment. Each annotated frame contains multiple instances (i.e. persons) with unique identifications, bounding boxes, and class/label information. The dataset and annotations can be used to train, validate and test the deep learning and computer vision based facial mask detection algorithms. Below are the details of dataset:

Total video frames: 4357
Total bounding boxes: 21941
Boxes with Mask (MW): 8306
Boxes without Mask (NM): 13635

Image frames: This folder contains 4357 video frames (.png). Each frame contains multiple instances.
Annotations: This folder contains the 4357 annotations (.xml) files for the above Image frames. Each .XML contains information about:

Person ID: Unique identification for each person in a frame
Bounding Box: The rectangular boundary around the person
Class Names: Mask (MW), no Mask (NM)

NOTE: Instances with mask and facing towards the camera are only labelled as being mask wearer (MW). All other subjects (i.e. instances) facing opposite to camera (i.e. backside towards the capturing device) are labelled as NM (i.e. without mask).