How to apply augmentation with parameters specific to each input image

[rhan93]

I have a dataset of 3D images of difference spacings. In an augmentation pipeline, I need to apply a resample transform such that the x&y spacing is the same as input and the z spacing is resampled to 3 mm, like tio.Resample([img.spacing[0], img.spacing[1], 3]). Can this be achieved in torchio?
In other words, in an augmentation transform list, can the resample parameters be dynamically changed according to the current image?

transforms = tio.Compose([tio.Resample(img.spacing[0], img.spacing[1], 3)])
dataset = tio.SubjectsDataset(subjects, transform=transforms)

[fepegar]

Hi, @rhan93. You could create your own transform for this:

import torchio as tio

class ResampleZ:
    def __init__(self, sz):
        self.sz = sz
    def __call__(self, subject):
        sx, sy, _ = subject.spacing
        resample = tio.Resample((sx, sy, self.sz))
        resampled = resample(subject)
        return resampled

image = tio.datasets.FPG().t1
transforms = tio.RandomBlur(), ResampleZ(4)
transform = tio.Compose(transforms)
transformed = transform(image)

[rhan93]

Thank you!