Hello

I compared the affine matrix after resampling, with different soft (mrtrix, sitk from torchio)
and I notice a small difference in the translation part of the affine when using nibabel function: nib.affines.rescale_affine()

I could find the same result if I modify nibabel code like that

Instead of

    centroid = nib.affines.apply_affine(affine, (shape -1) // 2)
    t_out = centroid - rzs_out @ ((new_shape -1) // 2)

I chage to

    centroid = nib.affines.apply_affine(affine, (shape -1) / 2)
    t_out = centroid - rzs_out @ ((new_shape -1) / 2)

it make sense to me to stay with float value for the centroid coordinate.

Cheers
Romain