int main(int argc, char* argv[])
{
try
{
// init comandline parser
TCLAP::CmdLine cmd("Command description message", ' ', "1.0");
TCLAP::ValueArg<string> argOutputFile ("o", "output", "Output file", true, "", "string");
TCLAP::ValueArg<string> argFilter ("f", "filter", "Filter files according to pattern", false, ".SIFTComparison.feat.xml.gz","string");
TCLAP::ValueArg<int> argPRECLUSTER ("p", "precluster", "Number of descriptors to select in precluster-preprocessing (0 = no preclustering)",false,0 ,"int");
TCLAP::ValueArg<int> argBOWSIZE ("b", "bowsize", "Size of the BoW Dictionary", false, 1000, "int");
TCLAP::SwitchArg argBinaryInput ("i", "binary", "Read descriptors from binary archives", false);
TCLAP::SwitchArg argVerbose ("v", "verbose", "Provide additional debugging output", false);
TCLAP::UnlabeledValueArg<string> dirArg ( "directory", "Directory containing files with extracted features", true, "directory","string");
cmd.add( argOutputFile );
cmd.add( argFilter );
cmd.add( argPRECLUSTER );
cmd.add( argBOWSIZE );
cmd.add( argBinaryInput );
cmd.add( argVerbose );
cmd.add( dirArg );
// parse arguments
cmd.parse( argc, argv );
// enable/disable verbose output
VerboseOutput::verbose = argVerbose.getValue();
VerboseOutput::println(string("train"), "Create BoW of size %d", argBOWSIZE.getValue());
TermCriteria tcrit;
tcrit.epsilon = 10;
tcrit.maxCount = 10;
tcrit.type = 1;
BOWKMeansTrainer bowtrainer(argBOWSIZE.getValue(),tcrit,1,KMEANS_PP_CENTERS);
VerboseOutput::println(string("train"), "Creating Visual Bag of Words");
string filter = argFilter.getValue();
if (argBinaryInput.getValue())
{
filter = ".SIFTComparison.descriptors.dat";
}
vector<string> files = getdir(dirArg.getValue(), filter);
int i = 1;
VerboseOutput::println(string("train"), "Reading features of directory '%s'", dirArg.getValue().c_str());
for (vector<string>::iterator filename = files.begin(); filename!=files.end(); ++filename)
{
VerboseOutput::println("train", "[%i of %i in directory '%s']", i++, files.size(), dirArg.getValue().c_str());
Mat descriptors;
stringstream filePathss;
filePathss << dirArg.getValue() << "/" << *filename;
string filePath = filePathss.str();
VerboseOutput::println(string("train"), string("processing file '" + filePath + "'"));
if (argBinaryInput.getValue())
{
loadDescriptorsFromBinaryArchives(descriptors, filePath);
}
else
{
loadDescriptorsFromOpenCVFilestorage(descriptors, filePath);
}
if ((descriptors.rows == 0) || (descriptors.cols == 0))
{
throw runtime_error("No Descriptors read for file: ");
}
VerboseOutput::println(string("train"), "%i descriptors loaded", descriptors.rows);
if ((argPRECLUSTER.getValue() > 0) && (argPRECLUSTER.getValue() < descriptors.rows - 100))
{
VerboseOutput::println(string("train"), string("pre-clustering"));
Mat labels;
Mat centers;
kmeans(descriptors,argPRECLUSTER.getValue(),labels,tcrit,1, KMEANS_PP_CENTERS, centers);
VerboseOutput::println(string("train"), "...add cluster centers of pre-clustering to bow");
bowtrainer.add(centers);
}
else
{
VerboseOutput::println(string("train"), "...add descriptors to bow");
bowtrainer.add(descriptors);
}
VerboseOutput::println(string("train"), "...current bow-size: %i", bowtrainer.descripotorsCount());
}
// calculate vocabulary
VerboseOutput::println(string("train"), string("Creating Vocabulary"));
if (bowtrainer.descripotorsCount() <= argBOWSIZE.getValue())
{
throw runtime_error("BoW size higher than number of loaded descriptors!");
}
Mat vocab = bowtrainer.cluster();
// output results to file
VerboseOutput::println(string("train"), string("Storing BoW to File"));
writeVocabularyToFile(vocab, argOutputFile.getValue());
VerboseOutput::println(string("train"), string("Finished"));
}
catch (exception &e) // catch any exceptions
{
cout << "\n";
cout << "*** Training aborted!\n";
cout << " Reason: " << e.what() << "\n\n";
exit(1);
}
return 0;
}
TCLAP::ValueArg<double> highThresholdArg( "t", "high", "High threshold value", false, 0.5, "" ); cmd.add( highThresholdArg ); /*---------------------------------------------------------------------*/ TCLAP::UnlabeledValueArg<std::string> inFilenameArg( "input", "Input image filename", true, "", "filename1" ); cmd.add( inFilenameArg ); TCLAP::UnlabeledValueArg<std::string> outFilenameArg( "output", "Output image filename", true, "", "filename2" ); cmd.add( outFilenameArg ); cmd.parse( argc, argv ); /***************************************************/ std::string inFilename = inFilenameArg.getValue(); std::string outFilename = outFilenameArg.getValue(); std::cout << "Loading file '" << inFilename << "' ..."; std::cout.flush(); M4D::Imaging::AImage::Ptr image = M4D::Imaging::ImageFactory::LoadDumpedImage( inFilename ); std::cout << "Done\n"; std::cout << "Initializing..."; std::cout.flush(); M4D::Imaging::PipelineContainer *container = NULL; FinishHook *hook = new FinishHook; M4D::Imaging::ConnectionInterfaceTyped< M4D::Imaging::AImage > *inConnection = NULL; M4D::Imaging::ConnectionInterfaceTyped< M4D::Imaging::AImage > *outConnection = NULL; M4D::Imaging::APipeFilter *filter = NULL; /*---------------------------------------------------------------------*/