Is running to generate training data for SVM.
Previously a trained SVM model works fine on a few test image, then decided to generate a large data to boost SVM performance.
However, the data set is 1.5Gb and svm-train from libsvm already crashed on that , seems the system does not support that much of memory use of libsvm , another half an hour wasted.
Will add svmprediction directly to the source, but will probably on CPU instead of GPU.
Do you have a smaller dataset to try? Or perhaps another machine?
ReplyDelete