I have trained a SVM model for detecting handwritten characters. On my desktop, I am able to save and load the model and its working perfectly. This is what I've used.
svm.save("./activ_hog_svm");
svm.load("./activ_hog_svm");
It works with both relative and absolute paths. However when I try to load the saved model in my Android app, the load command doesn't seem to work.
JNIEXPORT void JNICALL Java_co_activ_exp_MainActivity_loadSvm(JNIEnv* env, jobject, jstring svm_location)
{
const char *c_dir = env->GetStringUTFChars(svm_location, NULL);
LOGD("SVM location %s", c_dir);
// This prints /storage/emulated/0/Android/data/co.activ.exp/files/mounted/activ_hog_svm
// I have ensured that the file exists and has right permissions
svm.load(c_dir);
env->ReleaseStringUTFChars(svm_location, c_dir);
}
The model sits in assets folder. I also tried putting it directly at /sdcard but that didn't work either.
The error that shows up in logcat when I run a svm.predict later is
OpenCV Error: Bad argument (The SVM should be trained first) in CvSVM::predict, file /hdd2/buildbot/slaves/slave_ardbeg1/50-SDK/opencv/modules/ml/src/svm.cpp, line 21
Any help on how to fix this? Thanks.
