How to pre-process images during training?

[haimat]

I would like to blur my images before using them for training in Anomalib.
Is there a data loader or callback or something I could use in order to implement such a pre-processor?

[samet-akcay]

Would something like this work?

from torchvision import transforms
from anomalib.data import Visa

transform = transforms.Compose([
    transforms.Resize((224, 224)),  # Resize the image if needed
    transforms.GaussianBlur(kernel_size=(5, 5), sigma=(0.1, 2.0)),
    ...
])
datamodule = Visa(..., transform=transform, ...)

[haimat]

That looks great, thanks!