Hi Saygin.Let us consider your smallest class, the size of which will limit the maximal number of training/validation pixels. You can write:
Nb_Pixels_Smallest_Class = 115 pixels Ratio = 0.5Then, for EACH class, your training is made over about (0.5 x 115) = 57.5 pixels. Moreover, you have 4 classes, then you will train your model over a total number of (4 x 57.5) = 230 pixels, which approximately corresponds to your actual number of 232 training samples. Your validation data set is composed of 184 pixels corresponding to about (184./4) = 46 pixels for EACH class, which is quite close of the 57.5 pixels value of the training set. In fact both values are not strictly identical because the actual number of pixels over which the random sample selection is made (115 pixels) is too small to obtain similar values. However, you can consider that they are close enough to each other.
Please find below another example I gave as an answer over the otb-users mailing list concerning a similar issue but concerning both "sample.mt" and "sample.mv" parameters of the OTB Application otbapp_TrainImagesClassifier, which also have an infmuence of the actual number of samples randomly selected:
Considering Nb_Pixels_Smallest_Class = 430 pixels. Let us consider now an unbalanced ratio R = 0.1, (90% is used for Training and 10% for Validation). You will randomly select about 387 samples for T and 43 for V in EACH of your nbC classes if "sample.mt" = "sample.mv" = (-1).
If Ratio=0.1: (90% is used for T and 10% for V):If "sample.mt" = 100 < 430*0.9, you will use about 100 samples for T for each of your nbC classes. If "sample.mv" = 100 > 430*0.1, you will use about 43 samples for V for each of your nbC classes.
Then, to sum up, if you want to use the maximal available number of samples per class for T and V, you simply need to choose "sample.mt" = "sample.mv" = (-1). However, this maximal number of samples per class will be limited by the size of your smallest class. Then, you can set the ratio R according to your needs.
Best regards. Charles Peyrega Le 28/01/2014 12:43, saygin a écrit :
Hello,I am trying to train my .tif image with approximately 1500 pixels. I have 4 classes and one of them has the lowest training with 115 pixels. If I chose training and validation sample ratio 0.5, and using SVM OpenCV RBF with parameter optimization I get 232 training and 184 validation at log file. I am not very familiar with OTB and want to ask does not it use approximately half of all training pixels?Thanks -- -- 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/groups/opt_out.
-- <http://www.c-s.fr> *Charles PEYREGA*, /PhD/ Ingénieur d'études métier*Business Unit E-SPACE & Geo Information* <https://thor.si.c-s.fr/blogs/cs-blogs-business/>*- Département APPLICATIONS*
*CS Systèmes d'Information* Parc de la Grande Plaine - 5, Rue Brindejonc des Moulinais - BP 15872 31506 Toulouse Cedex 05 - FRANCE +33 561 17 65 14 - [email protected] -- -- 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/groups/opt_out.
<<inline: logo_cs.png>>
