I have seen that in many cases, specially with a sharp object, agisoft classifies most of a pair of images' matches as bad ones, and by seing them on the screen we can see that they are actually right. After many tryings with different values for the parameters, dividing in chunks, adding markers to the problematic photos, etc., I finally had to use a different software to compute the orientations.
This problem could be solved by allowing the user to mark the matches as good or bad, and afterward to run a "readjust" step: a computation of the images alignment based on the existing matches (which in addition has the advantage of being faster than the whole process of alignment).
Regards
Topic: Manually indicating good and bad matches (Read 4239 times)
