I found in line in BestOf2NearestMatcher, high confidence matches are set to zero confidence,
I also notice that in line in AffineBestOf2NearestMatcher, the line is commented.

In my experiment, I found with high confidence matches setting to zero, I lost some adjacent pictures, which overlap over 40%, which leads to blank in my final result. When I calculated the confidence outside the function and set the confidence > 3.0 's match to its actual value, I got correct result. And I suggest to comment the line line in BestOf2NearestMatcher
Is there anybody have same experience with me?