model.py

from keras import layers
from tensorflow import keras
from keras.models import Sequential
from keras.layers import Conv2D,Dropout,Flatten,Dense,MaxPooling2D, BatchNormalization

class MyModel(keras.Model):
    """
  Author: Trần Minh Hải
  Model được custom với số filter tăng lên:
    + Có 4 stage, mỗi satge có 2 conv2d layers và một maxpooling layers
    + Số filter qua mỗi stage sẽ được nhân 2 lên

  """
    def __init__(self, filter = 64, kernel = (3,3) ,  node = 64, outputs = 1):
        super(MyModel, self).__init__()
        self.filter = filter
        self.kernel = kernel
        self.node = node
        self.outputs = outputs
        
        self.stage1 = keras.Sequential([
                          layers.Conv2D(self.filter, kernel_size = self.kernel, activation='relu'),
                          layers.Conv2D(self.filter, kernel_size = self.kernel, activation='relu'),
                          layers.MaxPooling2D(pool_size=(2,2))
        ])
        self.stage2 = keras.Sequential([
                          layers.Conv2D(self.filter*2,kernel_size = self.kernel, activation='relu') ,
                          layers.Conv2D(self.filter*2,kernel_size = self.kernel, activation='relu'),
                          layers.MaxPooling2D(pool_size=(2,2))
        ])
        self.stage3 = keras.Sequential([
                          layers.Conv2D(self.filter*4,kernel_size = self.kernel, activation='relu',padding='same') ,
                          layers.Conv2D(self.filter*4,kernel_size = self.kernel, activation='relu', padding='same'),
                          layers.MaxPooling2D(pool_size=(2,2))
        ])
        self.stage4 = keras.Sequential([
                          layers.Conv2D(self.filter*8,kernel_size = self.kernel, activation='relu', padding='same') ,
                          layers.Conv2D(self.filter*8,kernel_size = self.kernel, activation='relu', padding='same'),
                          layers.MaxPooling2D(pool_size=(2,2), padding='same')
        ])
        self.outputs = keras.Sequential([
                                        layers.Flatten(),
                                        layers.Dense(self.node*64, activation = "relu"),
                                        layers.Dense(self.outputs, activation = "softmax")
        ])

    def call(self, inputs):
        x = self.stage1(inputs)
        x = self.stage2(x)
        x = self.stage3(x)
        x = self.stage4(x)
        x = self.outputs(x)
        return x