Face Mask Detection Using Deep Learning

EasyChair Preprint 10144

Abstract

Research suggests that COVID-19 facemasks degrade the performance of face detection technology. The purpose of this study is to quantify this effect and identify the ideal way to train face detection models when facemasks are present.

We tested the face detection capabilities of the model on both regular human faces and masked faces, recording the accuracy and recall rate for each group. Then we trained a new model, incorporating masked faces into the initial training dataset.

The adjusted model was tested to determine if the adjusted training set improved performance. This research will benefit machine learning researchers and data scientist who will train and utilize facial mask recognition models in the midst of COVID-19 and beyond.

The proposed technique is ensemble of one-stage and two-stage detectors to achieve low inference time and high accuracy.

In addition, we also propose a bounding box transformation to improve localization performance during mask detection. We explored the possibility of these models to plug-in with the proposed model so that highly accurate results can be achieved in less inference time.

Keyphrases: Face Mask, deep learning, detection

@booklet{EasyChair:10144,
  author    = {Wasudeo Rahane and Jayesh Kulkarni and Omkar Datir and Nikhil Kale and Viraj Kawade},
  title     = {Face Mask Detection Using Deep Learning},
  howpublished = {EasyChair Preprint 10144},
  year      = {EasyChair, 2023}}