|View Issue Details|
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0001370||Orfeo Toolbox (OTB)||General||public||2017-03-16 12:50||2017-04-05 10:08|
|Summary||0001370: Segmentation algorithm doesn't segment the whole image|
|Description||When running the otbcli_segmentation shell script, the algorithm doesn't perform the segmentation across the whole image. Sometimes I receive the error: |
2017 Mar 16 11:30:06 : Application.logger (INFO) Using watershed segmentation.
2017 Mar 16 11:30:06 : Application.logger (INFO) Large scale segmentation mode which output vector data
Computing watershed segmentation: 33% [**************** ]2017 Mar 16 11:30:27 : Application.logger (FATAL) The following error occurred during application execution : /home/mrashad/dashboard/otb/install-stable/include/ITK-4.10/itkWatershedSegmentTreeGenerator.hxx:437:
itk::ERROR: itk::watershed::SegmentTreeGenerator::MergeSegments:: An unexpected and fatal error has occurred. This is probably the result of overthresholding of the input image.
However, sometimes it reaches 100%, but still produces the same result.
I've tested this on the watershed and meanshift algorithms getting the same error both times.
I had this issue with Pleiades imagery on a much smaller scale, but it is a much larger issue when dealing with Worldview 3 imagery.
Is this a known issue, or is there a parameter I should be adding?
|Steps To Reproduce||wv3_segimg = "/home/williamray/Documents/Building_Recognition/Worldview3/Mexico/image_clip.tif"|
segments = "segments.shp"
wv3_segs = "wv3_segs.kea"
os.system("/home/williamray/OTB-5.8.0-Linux64/bin/otbcli_Segmentation -in %s -mode.vector.out %s -filter meanshift" %(wv3_segimg, segments))
os.system("gdal_rasterize -a DN -tr 2 2 -ot UInt32 -of KEA %s %s" %(segments, wv3_segs))
rastergis.populateStats(wv3_segs, True, True)
|Tags||No tags attached.|
I tried to reproduce your problem with WorldView2 and Pleiades images, but without success.
Could you give us your images where the problem is present (Pleiades or Worldview3) with a link to an host (since the limit on Mantis is of ~2MB) or even the steps to recreate your input data ?
|Unfortunately, due to licensing I can't share the imagery. I have played around with several parameters and am getting more stable results now. I believe it is a combination of the area of interest extent, the number of bands and the resolution of the imagery I'm using.|
It is unfortunate, but is it possible to reproduce this bug with images publicly available?
I know that its can take you some time to reproduce this bug with other data, but without that I am not able to fix the problem.
Nevertheless, I am glad that you have find a work around to perform full image segmentation.
|Changing to "feedback". We can't investigate unless able to reproduce.|
|2017-03-16 12:50||wray_qurus||New Issue|
|2017-03-16 12:50||wray_qurus||File Added: Segmentation.JPG|
||Status||new => acknowledged|
||Note Added: 0004072|
|2017-03-22 12:15||wray_qurus||Note Added: 0004076|
||Note Added: 0004078|
|2017-04-05 10:08||poughov||Note Added: 0004088|
|2017-04-05 10:08||poughov||Assigned To||=> poughov|
|2017-04-05 10:08||poughov||Status||acknowledged => feedback|