composition of Pad and Crop lead to a nifti header change

[romainVala]

🐛Bug

Not sure if it is a bug, sorry if I miss something: I want to do a padding (before model prediction) and then a crop on the prediction, but I notice a change in the nifti header, which make it as if there was a translation (if I save the output in a nifti file )

To reproduce

# Your code here
import torchio as tio
sub = tio.datasets.Colin27()

t = tio.Compose([tio.Pad(padding=10), tio.Crop(bounds_parameters=10)])
new_sub = t(sub)

print(new_sub.t1.affine)
print(sub.t1.affine)

# Your errors here

Expected behavior
the transformed image should have the save affine no ?

Actual behavior

TorchIO version

[fepegar]

Yes. It looks like a bug! I'll take a look.

[fepegar]

The problem seems to be in Crop.

[fepegar]

Fixed in v0.17.53. Thanks for reporting!

[romainVala]

thanks for correcting !