Ways to Wear a Mask or a Respirator (WWMR-DB)

Citation Author(s):
Antonio
Marceddu
Politecnico di Torino
Renato
Ferrero
Politecnico di Torino
Bartolomeo
Montrucchio
Politecnico di Torino
Submitted by:
Antonio Marceddu
Last updated:
Tue, 04/13/2021 - 11:28
DOI:
10.21227/8atn-gn55
License:
4
1 rating - Please login to submit your rating.

Abstract 

Mask detection is a task in where it is wanted to detect whether a person is wearing a mask or not. It seems like a simple problem, but some facets, such as the fact that people wear such masks and respirators in a multitude of ways, are often ignored and trivialized to the worn/not worn case.

 

The Ways To Wear a Mask or a Respirator Database (WWMR-DB) is a test database that can be used to compare the behavior of current mask detection systems with images that most closely resemble the real case. It consists of 1225 images divided into 8 classes, depicting the most common ways in which masks or respirators are worn:

- No Mask Or Respirator

- Mask Or Respirator Worn Correctly

- Mask Or Respirator Under The Nose

- Mask Or Respirator Under The Chin

- Mask Or Respirator Hanging From One Ear

- Mask Or Respirator On The Tip Of The Nose

- Mask Or Respirator On The Forehead

- Mask Folded Above The Chin

Each image has two label files: one in PascalVOC format and one in YOLO format. These files contain the bounding boxes of the face of the person, the mask or the respirator, and both.

 

Instructions: 

The database is ready-to-be-used and, due to its size, it was divided into three files:

- WWMR-DB - part 1.zip and WWMR-DB - part 2.zip files contain the database images;

- WWMR-DB - Labels.zip file contains the labels in PascalVOC and the YOLO format for each database image.

 

Due to the limitations imposed by the coronavirus, the database was created by asking volunteers for selfies through Google Forms. For this reason:

- number of images per class

- image quality

- intra-class differences

- rotation of the face

could also have great variations.

 

Google Forms are still open: if you want to contribute to the database, you can easily submit your images through the following links:

- Front photos (at 0 degrees): https://forms.gle/qLBjfCVhGoaJhnSo9

- Side photos (at 45 degrees): https://forms.gle/D3BuUQgjBLd6dqPj6

- Profile photos (at 90 degrees): https://forms.gle/yHgCAgcGJrfC7X4YA