Specify augmentation for different images

[rousseau]

First, thanks for this great tool !

I have a question regarding the data augmentation strategy. If for each subject, there is one image (T1). Is it possible to apply two different augmentations ?

For instance :
In a case of supervised denoising method,
training_transform_T1a = tio.Compose([spatial, noise])
training_transform_T1b = tio.Compose([spatial])
(the spatial deformations are the same for both images)

Or should we use the "include" option of the transform object and maybe specify 2 images for each subject ?

[fepegar]

Hi, @rousseau. I believe you can use include/exclude for this.

import torchio as tio
transform = tio.Compose([
    tio.RandomAffine(),
    tio.RandomBlur(std=(2, 3), include='t1'),
])
subject = tio.datasets.Pediatric(years=(4.5, 8.5))
subject.remove_image('pd')
subject.remove_image('mask')
subject.plot(show=False)
transformed = transform(subject)
transformed.plot()

[rousseau]

Thank a lot.

[fepegar]

No worries!