SVM classification and Feature extraction issues
Mantis Issue 280, reported by sdinot, assigned to cvalladeau, created: 2011-04-29
1/ SVN classification module can crash 2/ Pan Tex ans SFS textures of the Feature Extraction module are really slow.
See the snap shots in attachment.
1304323515 - C ValladeauYour troubles are all due to the complexity of the used algorithms. 1/ SVM Learning computes the hyperplan that separates all your classes. If you select numerous training ROIs (or big one), the process will take time. When the learining step is proceed, the "Learn" button is colored. As this process is done in the main thread, hard killing it, close the whole software. Maybe you can do the learning on a smaller image, save the generated SVM model and do the classification of your full image using this model. Furethemore, new the OTB-Applications classification chain can help you : http://wiki.orfeo-toolbox.org/index.php/Classification_chain.
2/ SFS and PanTex textures algorithms are too very complex (http://www.orfeo-toolbox.org/doxygen-current/classotb_1_1ScalarImageToPanTexTextureFilter.html#_details, http://www.orfeo-toolbox.org/doxygen-current/classSFSTexturesFunctor.html). We already have improved them and for instance, we don't have better implementation.
If you have