I am currently studying insightface and in particular this preprocess function.
It uses both scikit-image SimilarityTransform function (documentation) and openCV warpAffine function (documentation)
Is there a fundamental difference between the two or are they simply two different ways of calling the same mathematical transformation ?
A similarity transform is a special case of an affine transform, in which the shear is 0. (Only scale, rotation, and translation are allowed.) scikit-image also has an AffineTransform object.
The transform objects in scikit-image can be used both to estimate the transform, as pointed out by Piotr, but also to perform the transform, using skimage.transform.warp.