This stitching example
https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_feature2d/py_matcher/py_matcher.html
pipes the output of flann.knnMatch to cv2.drawMatchesKnn. Since the reference pages do not come close to properly documenting the output of knnMatch:
- https://docs.opencv.org/3.1.0/dc/de2/classcv_1_1FlannBasedMatcher.html
- https://docs.opencv.org/3.1.0/d4/de0/classcv_1_1DMatch.html
I decided to look at the implementation of drawMatchesKnn, but found something strange. In the python-code, the output of knnMatch is a list of lists of two DMatch objects:
matches[:3] == (
[[<DMatch 0x7ff57f981710>, <DMatch 0x7ff57f981130>],
[<DMatch 0x7ff57f981330>, <DMatch 0x7ff57f981290>],
[<DMatch 0x7ff57f981310>, <DMatch 0x7ff57f9816f0>]]
)
However drawMatchesKnn, to which matches is passed, these two line suggest it is a list of DMatches, instead of a list of lists of DMatches, since there is only one index [m] not two [m][n],
https://github.com/opencv/opencv/blob/05b15943d6a42c99e5f921b7dbaa8323f3c042c6/modules/features2d/src/draw.cpp#L209
Moreover, this loop for i,(m,n) in enumerate(matches): in the first link also seems to ignore the second DMatch n in each sublist.
What is this second DMatch in each item of matches? Why is matches a list of tuples of DMatches, and not a list of 'DMatch`es?
It seems as if I can ignore the second DMatch, yet I don't feel comfortable blindly doing so.