Hello,
I would like to use orfeo for a libsvm oneclass classification. However, I
get error messages that don't really help to see my mistake. Maybe someone
encountered a similar error.
Before I list my questions, I will explain what I did:
1. Generate training samples
- I created a polygone shapefile layer in QGIS and manually drew
polygons with class label 1 for the features, I would like to extract.
- result: Train_Samples_1Class.shp
2. Compute image second order statistics
- *otbcli_ComputeImagesStatistics -il namibia_orthoimage.tif -out
namibia_orthoimage_statistics.xml*
3. Train SVM classifier
- otbcli_TrainImagesClassifier -io.il namibia_orthoimage.tif -io.vd
Train_Samples_1Class.shp -io.imstat namibia_orthoimage_statistics.xml
-sample.mv 100 -sample.mt 100 -sample.vtr 0.5 -sample.vfn Class
-classifier
libsvm -classifier.libsvm.k linear -classifier.libsvm.c 1
*-classifier.libsvm.m
oneclass* -classifier.libsvm.opt 1 -io.out SVM_Model.txt
-io.confmatout SVM_ConfusionMatrix.csv
- This gives the error "*could not find paramter m*" when run as a
Pyhton script in the OSGeo4W Shell and "*option
-classifier.libsvm.m does not exist in the application*" when run
directly as the command-line listed above.
- Therefore, I tried it without the oneclass option and created
training samples with two classes (1: feature to be detected, 0 for
all
remaining polygons --> result: Train_Samples.shp) and changed the
above
code as follows:
- *otbcli_TrainImagesClassifier -io.il namibia_orthoimage.tif
-io.vd Train_Samples.shp -io.imstat namibia_orthoimage_statistics.xml
-sample.mv 100 -sample.mt 100 -sample.vtr 0.5 -sample.vfn Class
-classifier
libsvm -classifier.libsvm.k linear -classifier.libsvm.c 1
-classifier.libsvm.opt 1 -io.out SVM_Model.txt -io.confmatout
SVM_ConfusionMatrix.csv*
- This gives the error "*Input primary is required but not set*"
(ErrorI attached), when run as a Pyhton script in the OSGeo4W Shell
and "*Inconsistent
measurement vector size: Input Sample List size 3 Scale measurement
vector
size 0 Shift measurement vector size 0*" (Error II) when run
directly as the command-line listed above.
- I also attached the Python script, I refer to, to make it
clearer.
- When I alternatively try to do train the SVM classifier via the
Orfeo GUI in QGIS, I don't get any output or error message.
Questions:
- How do I correctly use the oneclass SVM classifier?
- Where is my error for training the two class SVM classifier?
- Does the output model file need to be .svm or .txt? I found different
versions in
https://www.orfeo-toolbox.org//CookBook/CookBooksu39.html#x58-910003.4.1
and in
https://www.orfeo-toolbox.org/CookBook/CookBooksu123.html#x154-9080004.8.10
Thanks a lot!
Sophie
--
--
Check the OTB FAQ at
http://www.orfeo-toolbox.org/FAQ.html
You received this message because you are subscribed to the Google
Groups "otb-users" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/otb-users?hl=en
---
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 [email protected].
For more options, visit https://groups.google.com/d/optout.
#### Train SVM Classifier ####
# Execute in OSGeo4W Shell:
# python Train_SVM.py namibia_orthoimage.tif Train_Samples.shp namibia_orthoimage_statistics.xml SVM_Model.txt SVM_ConfusionMatrix.csv
# otbcli_TrainImagesClassifier -io.il namibia_orthoimage.tif -io.vd Train_Samples.shp -io.imstat namibia_orthoimage_statistics.xml -sample.mv 100 -sample.mt 100 -sample.vtr 0.5 -sample.vfn Class -classifier libsvm -classifier.libsvm.k linear -classifier.libsvm.c 1 -classifier.libsvm.opt 1 -io.out SVM_Model.txt -io.confmatout SVM_ConfusionMatrix.csv
# Import required modules
#!/usr/bin/python
# -*- coding: utf-8 -*-
import otbApplication
from sys import argv
#### Main Part ####
# The following line creates an instance of the TrainImagesClassifier application
TrainImagesClassifier = otbApplication.Registry.CreateApplication("TrainImagesClassifier")
# The following lines set all the application parameters:
# Input Image List
TrainImagesClassifier.SetParameterStringList("io.il", argv[1])
# Input Vector Data List
TrainImagesClassifier.SetParameterStringList("io.vd", argv[2])
# Input XML image statistics file
TrainImagesClassifier.SetParameterString("io.imstat", argv[3])
TrainImagesClassifier.SetParameterInt("sample.mv", 100)
TrainImagesClassifier.SetParameterInt("sample.mt", 100)
# Training and validation sample ratio
TrainImagesClassifier.SetParameterFloat("sample.vtr", 0.5)
# Name of the discrimination field
TrainImagesClassifier.SetParameterString("sample.vfn", "Class")
# Classifier to use for the training
TrainImagesClassifier.SetParameterString("classifier","libsvm")
# SVM Kernel Type: Linear
TrainImagesClassifier.SetParameterString("classifier.libsvm.k","linear")
# Possible to change Kernel type later, then adapt gamma as well
# SVM Kernel Type: Gaussian radial basis function
# TrainImagesClassifier.SetParameterString("classifier.libsvm.k","rbf")
# SVM Model Type: Distribution estimation (One Class SVM)
# TrainImagesClassifier.SetParameterString("classifier.libsvm.m","oneclass")
# Cost parameter C
TrainImagesClassifier.SetParameterFloat("classifier.libsvm.c", 1)
# Parameters optimization
TrainImagesClassifier.SetParameterString("classifier.libsvm.opt","1")
# Output model
TrainImagesClassifier.SetParameterString("io.out", argv[4])
# Output confusion matrix
TrainImagesClassifier.SetParameterString("io.confmatout", argv[5])
# The following line execute the application
TrainImagesClassifier.ExecuteAndWriteOutput()
