How do I format the mask parameter for cv2.detail.BestOf2NearestMatcher apply2 function-CodePudding

I am looking to speed up some bundle adjustment code by specifying which images are overlapping. I created a function that places a grid of points in each image and then tests the pairwise overlap. Basically, it approximates the intersection area of the photos from the initial homography. Here is an example these 4 overlapping photos:

Ive been working off of this example https://github.com/amdecker/ir/blob/master/Stitcher.py

match_mask = np.zeros((len(features), len(features)), np.uint8)
    for i in range(len(data[0][1]) - 1):
        match_mask[i, i   1] = 1

    matches_info = matcher.apply2(features, match_mask)
.
.
.
 b, cameras = adjuster.apply(features, matches_info, cameras)

I can not find any documentation with regards to what format the match_mask needs to be. I would like to apply the mask such that only images with above some threshold say 50% overlap are used. If anyone could explain what the match_mask is that would be awesome!!

PS this is the docstring I found if it helps... finding it a bit cryptic doesnt give dimms or any details

"""
        apply2(features[, mask]) -> pairwise_matches
        .   @brief Performs images matching.
        .   
        .       @param features Features of the source images
        .       @param pairwise_matches Found pairwise matches
        .       @param mask Mask indicating which image pairs must be matched
        .   
        .       The function is parallelized with the TBB library.
        .   
        .       @sa detail::MatchesInfo
        """

CodePudding user response：

It looks like mask needs to be a NxN uint8 (boolean) matrix, N being the number of images, i.e. len(features).

If docs are unclear/vague/lacking, feel free to open an issue on the github about it. This function seems a good candidate for improvement.