ah I understand.
if your points contain no scaling, this will still give you a good result. you would only have to check if the rotation part of the matrix contains any scaling.
Why can’t I use the estimateAffinePartial2D？ It is the problem of my vision