Hi all,

Thanks for fixing pb 7.1. I'm looking forward to test it. 

For very Basic OBIA classification (on OBIA my personal use is mainly 
teaching, on a small areas , for the moment i make the student switch to 
scikit-learn for the classification after the attributes are produced in 

- As Jordi says, the ObjectsRadiometricStatistics in the OTB-contrib 
package or compiled from source works OK to derive radiometric attributes 
and basic shape attributes (elong, perim). However there is still the 
common issue of small vector areas compared to pixel size. In that case 
"nan" or empty attributes are produced. it would be good to make it more 
robust to small vector region.

- There are steps in the common classification pipeline that are still 
dependant on an image (XML for mean and variance of each feature in the OGR 
LayerClassifier, XML image stats file in the TrainVectorClassifier app) and 
prevent the use of all the radiometric attributes or extra attributes that 
could be calculated at object level.

- Maybe it would be interesting to adapt the SampleExtraction or 
TrainVectorClassifier so that the OBIA pipeline could benefit from the 
sample selection strategies.

I'm OK to spend time testing things on that if you want to validate the 
modification, or to add an OBIA exercice in the OTB workshop


Le vendredi 19 mai 2017 10:43:48 UTC+2, Julien Michel a écrit :
> Le 24/01/2017 à 17:38, Sébastien G. a écrit : 
> > *step 7- *otbgui_OGRLayerClassifier  that takes as input the previous 
> > shapefile and the image statistics and add a new "predicted" 
> > attributes in the shapefile. 
> > 
> >     pb7.1 : The OGRLayerClassifier only seem to accept SVM models. It 
> > worked for a SVM classifier but failed for a Random Forest with error 
> > messages related to itk SVM .... need to develop an otbapp 
> > VectorClassifier ? 
> Hi all, 
> I am just getting back to that thread, with a lot of activity and 
> comments. I just wanted to stress that pb7.1 is fixed in OTB 6.0 : you 
> know have access to all learning algorithm (including the unsupervised 
> shark K-Means we introduced in 6.0) in the TrainVectorClassifier app. 
> One step at a time ... There might also be in a near future a remote 
> module that allows to fix pb2.1. 
> Given all the insightful comments here, I think we could plan a working 
> group on this OBIA topic during next user days, to sum up what has been 
> said in the thread, add more ideas and write a good RFC to capture all 
> that. Then we could put it into action. 
> Thanks a lot for all your comments, 
> Regards, 
> Julien 
> -- 
> Julien MICHEL 

Check the OTB FAQ at

You received this message because you are subscribed to the Google
Groups "otb-users" group.
To post to this group, send email to otb-users@googlegroups.com
To unsubscribe from this group, send email to
For more options, visit this group at
You received this message because you are subscribed to the Google Groups 
"otb-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to