In feature extraction using SIFT, keeps only up to 1.5x of the
best min-num-features features (matching the behavior of HAHOG).
Very large images tend to result in VERY large number of features, which
slow down matching without adding any benefit. So this will speed up
the SFM step while using SIFT.
Makes the --matcher-order parameter kick-in only if the dataset is
not georeferenced. There's really no reason to apply matcher-order to
georeferenced datasets, where graph or distance matching works better.
min-num-featuresfeatures (matching the behavior of HAHOG). Very large images tend to result in VERY large number of features, which slow down matching without adding any benefit. So this will speed up the SFM step while using SIFT.--matcher-orderparameter kick-in only if the dataset is not georeferenced. There's really no reason to apply matcher-order to georeferenced datasets, where graph or distance matching works better.start-dev-env.shscript work on MacOS.