Detrimane if a person wears a mask or not using deep learning.
The goal is to use deep learning in general and convolutional neural network in particular convolutional neural networks in order to detranimine if the person in the picture wears a mask or not.
CNN is kind of the deep neural network, used mostly to analyze images. The goal of CNN is to reduce the dimensions of the picture to make analyze and prediction easier without losing important features. It uses filters, each one of them will represnt a certian features. In the start the features will be simple such as horizontal or diagonal edges, afterwards the filters will repsent more complex features such as wheels or eyes.
Wearing mask in this is not neccarly inuative. A person is wearing a mask if and only if 3 or more parts of his face are covered defined by this picture
The first part is specific functions for the dataset. The second part is the implementaions of the algorithms:
- model.py - Creation of the network and evlautes its performance and visualize on the test set.
- predict.py - Uses the model created to, given a path to new pictures outputs a csv file with the prections on those images.
The model is a 4 layers CNN that includes (on this order):
4 layers in the following structure:
Convolution with number of in channels is [3- RGB,16,32,64] (every i'th entry in the vector represent the in channel number of the i'th layer) and the number of out channels is [16,32,64,128] (every i'th entry in the vector represent the out channel number of the i'th layer). We can see that the in size of the i+1's layer equals the i's layer out size.
The batches are being normalized by the size of the out channels.
Then, it uses ReLu as an activation function and maxpooling the result with kernel of 2.It drops out 80% for regularization.
fc layer - The fully connected layer is a layer where each input is connected to all neurons. This layer helps provide and optimize the classification.
Finally, the model computes log soft max.
We saw that compared to other methods, it has the best impact on the result.
-
Open the terminal
-
Clone the project by:
$ git clone https://github.com/elaysason/Mask-Wearing-With-Deep-Learning.git
- Run the predict.py file by:
$ python predict.py /path/to/folder
The folder used as an input from predcit.py should hold images in the following format XXXXXX_Y.jpg: XXXXXX is an identificator for the image and Y is the label 0 for no mask and 1 for mask.