This week wasn't productive as it is more towards research on other models to solve an issue of faces with mask. No progression but alot of research.

 

There are alot of questions that goes through my mind in this sprint into how I am going to achieve a machine learning model to recognize with face. There are several things that I am thinking are as follows: 

​

- Using Eyes Biometric to recognize users identity (Iris)

- Using eyes and eye browns to recognize user

- Using simulation to predict user looks behind the mask

... and more

​

but there are many flaws as I was doing trial and error. It was a difficulty task for me as there aren't existing pretrained model for this. As for eyes biometric, the accuracy is 0.09. Face simulation, the accuracy is 0.2. I decided to take a step back to try to work on a Deep Convolutional Neural Network instead of trying to detect other features. In this model that I will try to achieve, in between I will be adding a layer to detect a mask to trigger the model to create points around the mask section. Otherwise, it will recognize the person face full feature if the mask is not wore.

​

These are the steps that I have planned to try to customize the machine learning model.

​

1. Create a custom FaceNet Model on how it trace the points of the mask

2. Create a model to detect mask or no mask

3. Face Alignment Processing using Single Shot Detection and as well as MTCNN

4. ResNet Model on face recognition (Either 50, 101 or 152) depends on how many layers will it take to have a better accuracy speed

​

Optimization

5. Adam Optimization

6. Decrease Weights to increase Recognition Speed

​

​

In between in each task, there will be data cleaning involve as well as other factors... Jiayous Team Asteria!