train_datagen = ImageDataGenerator(rescale=1. / 255)

train_generator = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode='rgb',
    class_mode=None)


validation_generator = test_datagen.flow_from_directory(
    validation_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    color_mode='rgb',
    class_mode=None)

input_img = Input(batch_shape=(None, img_width, img_width, 3))

#Encoder model

x = Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
encoded = MaxPooling2D((2, 2), padding='same')(x)

#decoder model

x = Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)
x = UpSampling2D((2, 2))(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), activation='relu')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(3, (3, 3), activation='sigmoid', padding='same')(x)

autoencoder = Model(input_img, decoded)
autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')
autoencoder.summary()


autoencoder.fit_generator(
    fixed_generator(train_generator),
    steps_per_epoch=nb_train_samples // batch_size,
    epochs=nb_epoch,
    validation_data=fixed_generator(validation_generator),
    validation_steps=nb_validation_samples // batch_size)

autoencoder.save_weights('anomaly-detection.h5')

#Getting a random image for prediction

im = cv2.resize(cv2.imread(filePath), (224, 224, 3))
im = im * 1. / 255
test_image[0, :, :, :] = im;

dec = autoencoder.predict(test_image)  # Decoded image
test_image = img[0]
dec = dec[0]
test_image = (test_image * 255).astype('uint8')
dec = (dec * 255).astype('uint8')
mse_value = mse(dec, test_image)

if mse_value > 2.74 : #Experimental threshold based on few iterations
       print('Image has an anomaly inside')