Insightface: why use skimage.transform for similarity transform

[zhenglaizhang]

I noticed the following code:

  tform = trans.SimilarityTransform()
  tform.estimate(dst, src)
  M = tform.params[0:2,:]
  #M = cv2.estimateRigidTransform( dst.reshape(1,5,2), src.reshape(1,5,2), False)

I looks like originally the transform M is estimated with opencv estimateRigidTransform, it may return None for bad data, is this the only reason to switch to skimage.transform.SimilarityTransform()?

[xmuszq]

version in cv2 is not robust. Do you have any other source to use other than skimage?

[safwankdb]

version in cv2 is not robust. Do you have any other source to use other than skimage?

Please check out petyr library. It is well suited for the task.