Teach on Udemy

Turn what you know into an opportunity and reach millions around the world.

Learn More

Your cart is empty.

Keep shopping

Deep Learning: masked face detection, recognition

Name: Deep Learning: masked face detection, recognition
Rating: 4.5 (49 reviews)

SSD face and facial mask detection, and train your own model to recognize faces even with masks

Created byJohnny Liao

Last updated 5/2022

English

What you'll learn

How to install Python, Tensorflow, Pycharm from scratch
How to create your own classification model
What's FaceNet
What's the difference between classification models and face recognition models
How to create your own FaceNet model by modifying the classification model
How to do the face alignment using SSD face detection
How to do the face alignment using MTCNN face detection
How to do the data cleaning
How to create masked face dataset
How to train your FaceNet model
What are training skills
How to implement training skills to train models effectively
How to perform the real time face detection, mask detection, and face recognition

Course content

18 sections • 54 lectures • 12h 9m total length

Environment installation38:47
System requirement
The overview of installation
Choose Nvidia GPU driver
Search your GPU driver
Nvidia GPU driver installation
Why we need virtual environments?
Learn more about Anaconda
Practice to create environments
Anaconda installation
Tensorflow installation test
Pycharm installation

Lecture_12:30
Show the packages that I use
Comparison of image processing packages
Lecture_27:13
Read images
Resize images
Change the color format
Transform to a gray image
Crop images
Save images
Display images
Lecture_38:26
How to read image paths from a root folder
How to select specific extension names
How to form 4 dimensional image data
How to distribute the dataset into train and test set7:02
How to write a function to distribute a dataset into training and test set

What is a classification model4:43
What is a classification model?
What’s inside the AI black box?
What is AI model training?
How to evaluate the model?
Recall, precision, and accuracy4:45
What is the accuracy?
In the binary system, what are TP, FP, TN, FN?
What is the recall?
What is the precision?
What is the accuracy in terms of positives and negatives?
Elements of a classification model17:11
Why uses the deep learning?
How CNN works?
Strides of CNN
Padding of CNN
Non-linearity
Activation functions
Pooling of CNN
Flatten
Fully connection
Dropout
Labels
One-hot encoding
Label dictionary example
Softmax function
Loss function: cross entropy
Optimizer functions
A model example: VGG
About the data set: No validation set in my tutorial

Introduction_18:53
•What’s Tensorflow?
•Why is Tensorflow difficult to understand?
•How to understand Tensorflow easily?
Introduction_218:45
How to use Tensorflow functions?
Why and how to set GPU resource?
Introduction_314:56
How to build deep learning models?
Use Tensorflow functions to create classic model: VGG16
How to define inputs?
Introduction of fashion mnist data set
What is the loss?
Set up the loss and the optimizer
Introduction_48:28
Read image paths, labels and record classes from the train folder
Read image paths and labels from the validation folder
An introduction of batch data, batch size, iterations, and epoch
How to calculate iterations?
How to do the training?
Introduction_513:12
How to evaluate the training?
Why do we need the validation data?
How to save the training weights?
What is transfer learning?
How to use PB file to do the inference?
Let's write codes29:25
Demonstrate how to write codes from scratch

Overview of classification model3:04
Build a classification model overview detailing class initialization, model initialization, and training, with inputs like dataset paths and label dictionary, and outputs such as weights, pb files, and logs.
Class initialization24:30
When you download the dataset(Mnist), please use the data distribution function(taught in Section 3) to divide the dataset into train and test set.
Model initialization34:28
Initialize a masked face detection model by constructing a simple resnet with input placeholders, a dropout-enabled fully connected layer, and softmax cross-entropy loss with an Adam optimizer.
Model train method39:19
Execute transfer learning for masked face detection and recognition, iterating through batches and epochs, adjusting the optimizer and batch size while evaluating training and test performance.
Save CKPT, PB files and log files16:22
Save training weights to CKPT and PB files
Save training results including the loss and the accuracy to the log file
The log file is a quick review if you want to know the parameters after 3 months

An introduction of Inception ResNet V128:32
Why can't we train FaceNet model with the simple ResNet model?
What's the lager model?
What is the standard evaluation method?
Create FaceNet model by modifying the classification model37:19
The test set input is replaced with another dataset
The inference model is replaced with a larger model
Add a standard evaluation method
Record all parameters
Check the training process is complete

Requirements

High school mathematics level
Basic Python and Tensorflow
Desktop or laptop with Windows and at least 6GB Nvidia GPU cards
A USB camera or a laptop camera

Description

Deep Learning of artificial intelligence(AI) is an exciting future technology with explosive growth.

Masked face recognition is a mesmerizing topic which contains several AI technologies including classifications, SSD object detection, MTCNN, FaceNet, data preparation, data cleaning, data augmentation, training skills, etc.

Nowadays, people are required to wear masks due to the COVID-19 pandemic.

The conventional FaceNet model barely recognizes faces without masks

Even the FaceID on iPhone or iPad devices only works without masks.

In this course, I will teach you how to train a model that works with masks.

In the final presentation, you will be able to perform the real time face detection, face mask detection, and face recognition, even with masks!

Windows is the operating system so you don't need to learn Linux first.

Having Python and Tensorflow knowledge are required.

In my tutorials, I would like to explain difficult theories and formulas by easy concepts or practical examples.

Model training always takes a lot of time.

Take this project as an example, it needs more than 400,000 images to train.

I will offer training skills to speed up the training process.

These training skills can be not only applied in face recognition but also in your future projects.

All lectures are spoken in plain English.

If you feel my speaking pace is quite slow, you can use the gear setting to speed up.

If you don't want to train the model by yourself, the source code and trained weight files are included!

Besides the training steps, this is also a highly integrated application.

Achievement from the topic, skills grow from the project. I hope you enjoy the fun of AI.

Who this course is for:

Those who have Python basics tend to learn Deep Learning or Face Recognition
Any engineers who want to level up in Deep Learning

Deep Learning: masked face detection, recognition

What you'll learn

Explore related topics

Course content

Set up the environment1 lecture • 39min

Jupyter notebook coding environment2 lectures • 5min

Image process4 lectures • 25min

Classification model explanation3 lectures • 27min

Tensorflow introduction and quick guide6 lectures • 1hr 34min

Write a classification class program5 lectures • 1hr 58min

FaceNet concepts2 lectures • 14min

Create FaceNet model2 lectures • 1hr 6min

Face alignment of CASIA dataset using SSD face detection3 lectures • 29min

Face alignment of CASIA dataset using MTCNN1 lecture • 26min

Requirements

Description

Who this course is for: