c++ - Recommended values for OpenCV SVM parameters -


any idea on recommended parameters opencv svm? i'm playing letter_recog.cpp in opencv sample directory, however, svm accuracy poor! in 1 run got 62% accuracy:

$ ./letter_recog_modified -data /home/cobalt/opencv/samples/data/letter-recognition.data  -save svm_letter_recog.xml -svm  database /home/cobalt/opencv/samples/data/letter-recognition.data loaded. training classifier ... data.size() = [16 x 20000] responses.size() = [1 x 20000]  recognition rate: train = 64.3%, test = 62.2%

the default parameters are:

model = svm::create(); model->settype(svm::c_svc); model->setkernel(svm::linear); model->setc(1); model->train(tdata);

setting trainauto() didn't help; gave me weird 0 % test accuracy:

model = svm::create(); model->settype(svm::c_svc); model->setkernel(svm::linear); model->trainauto(tdata);

result:

recognition rate: train = 0.0%, test = 0.0%

update using yangjie's answer:

$ ./letter_recog_modified -data /home/cobalt/opencv/samples/data/letter-recognition.data  -save svm_letter_recog.xml -svm database /home/cobalt/opencv/samples/data/letter-recognition.data loaded. training classifier ... data.size() = [16 x 20000] responses.size() = [1 x 20000]  recognition rate: train = 58.8%, test = 57.5%

the result no longer 0% accuracy worse 62% earlier.

using rbf kernel trainauto() worst?

$ ./letter_recog_modified_rbf -data /home/cobalt/opencv/samples/data/letter-recognition.data  -save svm_letter_recog.xml -svm database /home/cobalt/opencv/samples/data/letter-recognition.data loaded. training classifier ... data.size() = [16 x 20000] responses.size() = [1 x 20000]  recognition rate: train = 18.5%, test = 11.6%

parameters:

    model = svm::create();     model->settype(svm::c_svc);     model->setkernel(svm::rbf);     model->trainauto(tdata);

i debugged sample code , found reason.

the responses mat of ascii code of letters.

however, predicted labels returned svm trained svm::trainauto ranging 0-25, correspond 26 classes. can observed looking @ <class_labels>...</class_labels> in output file svm_letter_recog.xml.

therefore in test_and_save_classifier, r = model->predict( sample ) , responses.at<int>(i) apparently not equal.

i found if use svm::train, class labels 65-89 instead, why can normal result @ first.

solution

i not sure whether bug. if want use svm::trainauto in sample now, can change

test_and_save_classifier(model, data, responses, ntrain_samples, 0, filename_to_save);

in build_svm_classifier to

test_and_save_classifier(model, data, responses, ntrain_samples, 'a', filename_to_save);

update

trainauto , train should have same behavior in class_labels. problem due bug fix before. have created pull request opencv fix problem.


-

Comments

Post a Comment

Popular posts from this blog

OpenCV OpenCL: Convert Mat to Bitmap in JNI Layer for Android -

there several posts converting mat bitmap using utils.mattobitmap() function. i'm assuming function can called in java layer after importing utils class. i want transfer data memory address pointed uint32_t* bmpcontent; in code below. jniexport void jnicall java_com_nod_nodcv_nodcvactivity_runfilter( jnienv *env, jclass clazz, jobject outbmp, jbytearray indata, jint width, jint height, jint choice, jint filter) { int outsz = width*height; int insz = outsz + outsz/2; androidbitmapinfo bmpinfo; if (androidbitmap_getinfo(env, outbmp, &bmpinfo) < 0) { throwjavaexception(env,"gaussianblur","error retrieving bitmap meta data"); return; } if (bmpinfo.format != android_bitmap_format_rgba_8888) { throwjavaexception(env,"gaussianblur","expecting rgba_8888 format"); return; } uint32_t* bmpcontent; if (androidbitmap_lockpixels(env, outbmp,(void**)&bmpcontent)...
Read more

android - org.xmlpull.v1.XmlPullParserException: expected: START_TAG {http://schemas.xmlsoap.org/soap/envelope/}Envelope -

i trying hit soap api in application login purpose , getting above mentioned error , can 1 tell me going wrong , first time soap apis. here code hit soap :- vector<object> args = new vector<object>(); hashtable<string, object> hashtable = new hashtable<string, object>(); hashtable.put("email", "mss.siddhart28@gmail.com"); hashtable.put("password", "siddharth"); args.add(hashtable); soapserializationenvelope envelope = new soapserializationenvelope( soapenvelope.ver11); envelope.dotnet = true; envelope.xsd = soapserializationenvelope.xsd; envelope.enc = soapserializationenvelope.enc; (new marshalhashtable()).register(envelope); httptransportse androidhttptransport = new httptransportse( "http://tempuri.org/", 15000); androidhttptransport.debug = true; string result = ""; try { soapobject request = new soapobject("http:...
Read more

python - How to remove the Xframe Options header in django? -

i have made page has iframe . inside iframe want show multiple different links article facebook, or news, or youtube video or other possible url. but, due xframe header, unable so. referred following link: https://docs.djangoproject.com/en/1.8/ref/clickjacking/ , django xframeoptionsmiddleware (x-frame-options) - allow iframe client ip but didn't help. my settings.py file's middleware_classes is: middleware_classes = ( 'django.contrib.sessions.middleware.sessionmiddleware', 'django.middleware.common.commonmiddleware', 'django.middleware.csrf.csrfviewmiddleware', 'django.contrib.auth.middleware.authenticationmiddleware', 'django.contrib.auth.middleware.sessionauthenticationmiddleware', 'django.contrib.messages.middleware.messagemiddleware', 'django.middleware.clickjacking.xframeoptionsmiddleware', ) from http://django-secure.readthedocs.org/en/latest/middleware.html , found using ...
Read more