Poor performance of ImportReference

[luroth]

Hi,

I'm using PhotoScan Professional in a fully integrated Python script workflow to process aerial images. For precise georeferencing after a rought alignment based on GPS coordinates, I'm using the non-coded GCP detection feature of PhotoScan with a very relaxed tolerance, followed by a reference import that assigns coordinates to the nearest detected GCP and discards all others:

Code: [Select]

chunk.detectMarkers(PhotoScan.TargetType.CrossTarget, tolerance=1)
chunk.loadReference(_settings['campaign']["gcp_file"],
   PhotoScan.ReferenceFormatCSV,
   columns='nxyz', delimiter=',',
   ignore_labels=True, threshold = 0.4)

This code performs terribly slow if a lot of (false) GCPs were detected and had to become discarded. My 3.5 GHz 40 core CPU is on 3 % utilization, but the step takes up to 12 h. Could you please fix this performance issue, e.g. by allowing multiprocessing?

[Alexey Pasumansky]

Hello luroth,

How many markers are detected before you call import reference command?