CNN_model.py

kernel_size = 3
dropout = .5
activation_func = "relu"
input_shape = (54, 50, 3)

conv__filters_1 = 32
conv__filters_2 = 48
conv__filters_3 = 64
density_units_1 = 256
density_units_2 = 64
epochs          = 50

model = Sequential([
    # Conv layer #1
    Conv2D(
        filters = conv__filters_1, 
        kernel_size = kernel_size + 4, 
        activation  = activation_func, 
        input_shape = input_shape, #input layer
        padding     = "same"
    ),
    Conv2D(filters = conv__filters_1, kernel_size = kernel_size + 4, activation = activation_func, padding = "same"),
    MaxPooling2D(pool_size=(2,2)),
    Dropout(dropout/2),

    # Conv layer #2
    Conv2D(filters = conv__filters_2, kernel_size = kernel_size + 2, activation=activation_func, padding = "same"),
    Conv2D(filters = conv__filters_2, kernel_size = kernel_size + 2, activation = activation_func, padding = "same"),
    MaxPooling2D(pool_size=(2,2)),
    Dropout(dropout/2),

    # Conv layer #3
    Conv2D(filters = conv__filters_3, kernel_size = kernel_size, activation=activation_func, padding = "same"),
    Conv2D(filters = conv__filters_3, kernel_size = kernel_size, activation = activation_func, padding = "same"),
    MaxPooling2D(pool_size=(2,2)),
    Dropout(dropout/2),

    # Dense layer #1
    Flatten(),
    Dense(density_units_1, activation=activation_func),
    Dropout(dropout),

    # Dense layer #2
    Dense(density_units_2, activation=activation_func),
    Dropout(dropout),

    # Output layer
    Dense(1, activation='sigmoid')
])

model.compile(
    loss      = 'binary_crossentropy',
    optimizer = RMSprop(lr=0.0001, decay=1e-6),
    metrics   = ['binary_accuracy']
)