I am trying to build a CNN model in which it would recognize whether it is my face or not my face for educational purposes. However, as soon as I ran model.fit() and saw the training process, I immidiately realized that the model is suffering from a severe overfitting issue (in this case accuracy was close to 1 but val_acc remained at 0.4849). I dont know whether it is my mistake in choosing the right learning rate, epochs and batch size, or the model is too complex, or i just dont have enough data.

This is the model, I will be happy to provide additional code if necessary. Thank you!

size = 224
def create_model():
    model = Sequential([
        Input(shape=(size,size,3)),
        # First Hidden Layer
        Conv2D(32, (3,3), activation='relu', kernel_regularizer=tensorflow.keras.regularizers.l2(0.01), bias_initializer=RandomNormal()),
        Dropout(0.3),
        MaxPooling2D((2,2)),
    
        # Second Hidden Layer
        Conv2D(64, (3,3),1,activation='relu',  kernel_regularizer=tensorflow.keras.regularizers.l2(0.01)),
        BatchNormalization(),
        Dropout(0.3), 
        MaxPooling2D((2,2)),
    
        # Third Hidden Layer
        Conv2D(128, (3,3),1, activation='relu'),
        MaxPooling2D((2,2)),
    
        # Flatten the model
        Flatten(),
    
        Dropout(0.3), # Randomly reduces the data by 50% to prevent overfitting
        BatchNormalization(),
    
        # Fully Connected Layer
        Dense(256, activation='relu', kernel_regularizer=tensorflow.keras.regularizers.l2(0.01)),  # Add L2 regularization
        Dropout(0.3),
        Dense(1,activation='sigmoid') # Produces a probability output for binary classification
    ])
    
    prev_weight = model.layers[0].get_weights()
    if len(prev_weight) == 2:
        weight_mat, bias = prev_weight
        print("Current weight is {}".format(weight_mat))
        print("Current bias is {}".format(bias))
        random_bias = np.random.randn(* bias.shape)
        new_weight = [weight_mat * 0.1, random_bias]
        model.layers[0].set_weights(new_weight)
    else:
            raise ValueError("NGU VL!!!!")
    
    
    model.compile(loss="binary_crossentropy", optimizer=optimizer, metrics=["accuracy", ])
    return model