Overview / Usage

This project on Face mask Detection is completely based on Himanshu Tripathi's work on Face Mask Detection for COVID-19.

A pre-trained model called 'mobilenet' from ml5.js has been used for the implementation of this Deep Learning project wherein the principles of Transfer Learning has been used to train the model through new images. A JS client-based library called p5.js, which is based on the core principles of processing, has been used for creating the basic graphic layout ; and for deployment of Machine Learning in the web browser I have used ml5.js, which is fundamentally a HLL to TensorFlow.js.

Questionnaire related to this Project

1. When does the detection API detect that the person is not wearing a mask?
   So for this project to work, an image capturing device is required which gives us a continuous feedback overlay. If the person in front of the camera, is detected of not wearing a mask then it shoots us a warning of "Mask Off" in the web browser. So to answer this question, the Face Mask Recognition works in real time and gives us instantaneous warnings of its detection.

2. Whether there is any need for a camera to be pointed to the face for detecting the same.
   No, there is no need for the camera to be pointed directly at the face for the detection to be working. In my project I have trained the classifier to recognize whether or not I'm wearing a mask by pointing my face towards the camera. But it totally depends on how you train the classifier to classify your images.

3. Whether it can generate a voice or sound alarm in case of detection.
   In this project of mine I haven't triggered any voice or sound alarm in case of detection. But I'm sure that it is possible. Not only being limited to sound alarm, we can also send an auto generated e-mail to respective authorities in case it detects someone not wearing a mask. But then again I haven't implemented that into my project yet but am looking forward to doing this.

Methodology / Approach

Transfer Learning, so the basic idea is to train the mobilenet library for feature extraction and train the classifier using our examples.

After extraction of features from the new images inputted, a Neural network program trains the images

Transfer learning is the process of taking a pre trained model and disconnect it from the pre-trained labels and reconnect it to our own labels.

It is possible to save and load training data in ml5.js , and we go about doing this with classifier.save() and the classifier.load() function

When we save our transfer learning data into our local PC, two files get downloaded:

1. model.weights.bin : This file basically contains all the numbers and computational data of the Neural Network working behind the model. 'bin' in there stand for binary format which roughly contains millions numbers and connections b/w lot of hidden layers.
2. model.json : The JSON file basically is a text file of how the model is configured.

Technologies Used

Deep learning frameworks: Tensorflow.js, ml5.js, p5.js

Repository

https://github.com/killerrings/Face-Mask-Detection

Other links

• Bio about Project