Added AdaLAM wrapper.

[mihaidusmanu]

Some results on South-Building

SIFT + MNN

    num_reg_images = 128
    num_cameras = 1
    num_points3D = 40544
    num_observations = 215421
    mean_track_length = 5.31326
    mean_observations_per_image = 1682.98
    mean_reprojection_error = 0.930621
    num_input_images = 128

SIFT + AdaLAM

   num_reg_images = 128
   num_cameras = 1
   num_points3D = 38752
   num_observations = 214480
   mean_track_length = 5.53468
   mean_observations_per_image = 1675.62
   mean_reprojection_error = 0.949959
   num_input_images = 128

Extra

• Bugfix in GV if image has no keypoints

TODO

• Add support for features without ori / scale (by setting orientation_difference_threshold and scale_rate_threshold to None)

[mihaidusmanu]

@cavalli1234 for double checking

[cavalli1234]

Looks all good!

• Just consider that, in case confidence might be useful, an estimate of it can be extracted from AdaLAM (in adalam/core.py, function adalam_core, variable conf). It is the ratio between the local inliers found divided by the expected number of random false positives you would get from outliers only in the same local neighborhood, so 1 means insignificant, and goes up unbounded.