Hi,
After creating a CANUPO classifier using training data, you can start classififying a point cloud. In the option for classifying the point cloud ('CANUPO Classify'), one can specify to 'use active SF to locally refine the classification'. However, in the paper by Brodu and Lague I cannot find information about this. How does this refinement work? Especially because the classifier is already trained..
Thank you in advance!
CANUPO classification, refinement using active SF option
Re: CANUPO classification, refinement using active SF option
The tooltip says "Try to classify points with a low confidence based on the local SF values".
And when looking at the code, it seems that for points with a low confidence value (= according to the threshold above), Canupo will consider the scalar value and the classification of the neighbor points. Based on that, it will try to determine if the point should be more in one class or the other (by considering statistics on the scalar valus - see https://github.com/CloudCompare/CloudCo ... ss.cpp#L52).
So I guess you need a pretty 'contrasted' SF.
And when looking at the code, it seems that for points with a low confidence value (= according to the threshold above), Canupo will consider the scalar value and the classification of the neighbor points. Based on that, it will try to determine if the point should be more in one class or the other (by considering statistics on the scalar valus - see https://github.com/CloudCompare/CloudCo ... ss.cpp#L52).
So I guess you need a pretty 'contrasted' SF.
Daniel, CloudCompare admin