I have a problem with Keras where I am training two separate models one after the other. However, the second model seems to keep using the training data or states of the first model.

image_size = 150
    input_shape = (image_size, image_size, 3)

    pre_trained_model = VGG16(input_shape=input_shape, include_top=False, weights="imagenet")
        
    for layer in pre_trained_model.layers[:15]:
        layer.trainable = False

    for layer in pre_trained_model.layers[15:]:
        layer.trainable = True
        
    last_layer = pre_trained_model.get_layer('block5_pool')
    last_output = last_layer.output
        
 ........

    model = Model(pre_trained_model.input, x)

    model.compile(loss='binary_crossentropy',
                optimizer=optimizers.SGD(learning_rate=1e-4, momentum=0.9),
                metrics=['acc'])

    model.summary()
    return model

Init:

model = create_model()

Datagen:

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,)

train_generator = train_datagen.flow_from_directory(
    data_path+"/training_set",
    target_size=(150, 150),
    batch_size=5,
    class_mode='binary'
)

test_datagen = ImageDataGenerator(rescale=1./255)

validation_generator = test_datagen.flow_from_directory(
        data_path+"/test_set",
        target_size=(150, 150),
        batch_size=5,
        class_mode='binary')

Training:

history = model.fit(
    train_generator,
    steps_per_epoch=70,  # Anzahl der Batches pro Epoche
    epochs=20,
    validation_data=validation_generator,
    validation_steps=20  # Anzahl der Validierungsbatches
)

After the first training i Initalize everything the same:

model1 = create_model()

…….

history1 = model1.fit(
    train_generator,
    steps_per_epoch=70,  # Anzahl der Batches pro Epoche
    epochs=20,
    validation_data=validation_generator,
    validation_steps=20  # Anzahl der Validierungsbatches
)

although I use

tf.keras.backend.clear_session() 

and instantiate each model in a separate function. I am on the second epoch on an acc of 1.00.

How can I solve this?