CODEWITHSUNDEEP
pythontensorflowkerasneural-network
Petra Svobodova
Petra Svobodova

Reputation: 11

Add a custom preprocessing function in ImageDataGenerator for gaussian blurring

I am using Image Data Generator to extend my dataset.

train_generator = train_datagen.flow_from_directory(
        path_to_train,
        target_size=(150, 150),
        batch_size=32,
        class_mode = "sparse",
        color_mode="grayscale")

and then

history = model.fit(
        train_generator,
        steps_per_epoch=20,
        epochs=100
        )

The problem is that I want to apply some methods to the input images. I want to use Gaussian bluring and etc. But how can I apply it when I directly load it and use model fit.

Can you help me please? Thanks for you advices.

Upvotes: 0

Views: 1447

Answers (1)

Nicolas Gervais
Nicolas Gervais

Reputation: 36684

You can do this with tfa.image.gaussian_filter2d:

@tf.function
tfa.image.gaussian_filter2d(
    image: tfa.types.TensorLike,
    filter_shape: Union[List[int], Tuple[int], int] = [3, 3],
    sigma: Union[List[float], Tuple[float], float] = 1.0,
    padding: str = 'REFLECT',
    constant_values: tfa.types.TensorLike = 0,
    name: Optional[str] = None
) -> tfa.types.TensorLike

You can pass this function in ImageDataGenerator:

img_gen = ImageDataGenerator(
    proprocessing_function=tfa.image.gaussian_filter2d
)

Upvotes: 3

Related Questions