As for this week, I managed to get the model to recognize faces with facial mask with a better accuracy. I have added a Euclidean distance function to calculate the distance between two images. Making use of numpy argsort function to get the index of minimum distance. If the index is correct and the distance value is smaller than the treshold(0.8) which the face matching is correct. On top of that, this helps for better face cropping image processing to get rid of any unwanted data.

 

With the face with mask datasets that I have made, I will be using it to check the accuracy the custom model that I have trained previously to see accuracy outcome of face with facial mask. The number of images is 10,000 so it took quite awhile. Initially the accuracy reaches 0.6 but after tuning and optimizing the model, it managed to hit an accuracy of 0.99 which indeed seems very promising. I hope that I did not accidentally do any overfitting but no sign of overfitting. 

​

​

Side note what is ED?

Euclidean Distance represents the shortest distance between two points. Most machine learning algorithms including K-Means use this distance metric to measure the similarity between observations.

Formula = Sqrt((nth1 - nth2) ^ 2)