Backend Progress

[e-shen2022]

Read comments for line by line explanations

1. Import Packages

• Numpy to save image data in arrays
• Tensorflow and keras to create and train model

2. Train data

• Resize image input to fit dimensions for NN training
• Specify number of images per batch, and how many batches model will train
• Converts images to grayscale

3. Creating the CNN model

• Sequential model where layers increase in complexity of what pattern can recognize
• Adds convolutional layers to detect local patterns
• Add pooling layers to downsample data to look at small areas of image
• Dropout layer to regularize data and prevent overfitting
• Flatten layer to reshape to 1D array to prepare for connecting all layers
• Dense layer to connect all layers, last layer specifies 7 neurons which is 7 different possible emotion outputs

4. Compiling the model for training

• Prepare model for training by defining how it will measure loss, update its weights, and evaluate its prediction performance
• Save learned parameters in model.h5 for graphical user interface