I haven't seen this in any issued tagged "python bindings", so I'll give it a go...

The "pythonic" thing to do would be accepting lists/tuples where arrays are expected, and accepting lists/arrays where tuples are expected.

In one instance, an error message asking for an UMat (when the API wants a numpy array) has just confused a newbie in the IRC channel:

>>> contour = [[0,0], [10,0], [10,10], [0,10]]
>>> cv.pointPolygonTest(np.array(contour), (5,5), True)
5.0
>>> cv.pointPolygonTest(contour, (5,5), True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: Expected cv::UMat for argument 'contour'

Another slight impedance mismatch I see sometimes is the new style getargs format but argument is not a tuple error message when the bindings expect a tuple, to be converted into a cv::Size or similar C++ API type. That too could be made more accomodating.

I have no idea how OpenCV generates its python bindings. It used to be some custom code generation. Maybe someone would like to take a shot at it. With some discussion and pointers, this might even become suitable to work on for beginners?