mzur
commented
6 years ago In [1] they use DBSCAN clustering to merge point and line annotations:
To obtain a consensus on symmetry GTs computation ally, we first combine human perceived symmetry labels with an automated clustering algorithm [17]. The basic idea is to capture the exponential divergence in the nearest labeled symmetry pair distribution, use that as the minimum distance τ between neighbors and the number of required human labels as the minimum number of neighbors, and finally input both to DBSCAN [14], a method for Density-Based Spatial Clustering of Applications with Noise (the winner of the test-of-time award in 2014). The τ for rotation symmetry perception is 5 pixels, i.e. two symmetry labels within τ are considered to be labeling the same perceived symmetry [17].
With multiple labels per annotation and the possibility to conduct annotation sessions where the users don't see annotations of other users, reports may convey a false impression of the abundance of objects in transects. With the basic annotation report, for example, the count of annotation labels may no longer reflect the count of actual objects.
Implement an option to automatically count multiple annotations / annotation labels as one based on specific parameters:
Possibly related to this is the option to add "quality" or "certainty" restrictions like "only annotations that were merged from at least x actual annotations" or "only annotations with labels with a confidence higher than x".