QString NewFormWidget::itemToTemplate(const QTreeWidgetItem *item, QString *errorMessage) const
{
const QSize size = templateSize();
// file name or string contents?
const QVariant templateFileName = item->data(0, TemplateNameRole);
if (templateFileName.type() == QVariant::String) {
const QString fileName = templateFileName.toString();
// No fixed size: just open.
if (size.isNull())
return readAll(fileName, errorMessage);
// try to find a file matching the size, like "../640x480/xx.ui"
const QFileInfo fiBase(fileName);
QString sizeFileName;
QTextStream(&sizeFileName) << fiBase.path() << QDir::separator()
<< size.width() << QLatin1Char('x') << size.height() << QDir::separator()
<< fiBase.fileName();
if (QFileInfo(sizeFileName).isFile())
return readAll(sizeFileName, errorMessage);
// Nothing found, scale via DOM/temporary file
QString contents = readAll(fileName, errorMessage);
if (!contents.isEmpty())
contents = qdesigner_internal::WidgetDataBase::scaleFormTemplate(contents, size, false);
return contents;
}
// Content.
const QString className = item->data(0, ClassNameRole).toString();
QString contents = qdesigner_internal::WidgetDataBase::formTemplate(m_core, className, formName(className));
if (!size.isNull())
contents = qdesigner_internal::WidgetDataBase::scaleFormTemplate(contents, size, false);
return contents;
}
virtual cf_tracking::CfTracker* parseTrackerParas(TCLAP::CmdLine& cmd, int argc, const char** argv)
{
cf_tracking::DsstParameters paras;
TCLAP::SwitchArg debugOutput("v", "debug", "Output Debug info!", cmd, false);
TCLAP::SwitchArg originalVersion("", "original_version", "Use the original parameters found in the DSST paper. Performance is close, "
"but differences do still exist!", cmd, false);
TCLAP::ValueArg<int> templateSize("", "para_template_size", "template size", false,
paras.templateSize, "integer", cmd);
TCLAP::ValueArg<double> padding("", "para_padding", "padding around the target", false,
paras.padding, "double", cmd);
TCLAP::ValueArg<double> outputSigmaFactor("", "para_output_sigma_factor", "spatial bandwitdh of the target",
false, paras.outputSigmaFactor, "double", cmd);
TCLAP::ValueArg<int> cellSize("", "para_cell_size", "cell size of fhog", false, paras.cellSize, "integer", cmd);
TCLAP::ValueArg<double> lambda("", "para_lambda", "regularization factor", false, paras.lambda, "double", cmd);
TCLAP::ValueArg<double> interpFactor("", "para_interpFactor", "interpolation factor for learning",
false, paras.learningRate, "double", cmd);
TCLAP::ValueArg<double> scaleSigmaFactor("", "para_scale_sigma_factor", "spatial bandwitdh of the target(scale)",
false, paras.scaleSigmaFactor, "double", cmd);
TCLAP::ValueArg<double> scaleStep("", "para_scale_step", "scale_step", false, paras.scaleStep, "double", cmd);
TCLAP::ValueArg<int> scaleCellSize("", "para_scale_cell_size", "cell size of fhog (scale filter)", false, paras.scaleCellSize, "integer", cmd);
TCLAP::ValueArg<int> numberOfScales("", "para_scale_number", "number of scale steps", false, paras.numberOfScales, "integer", cmd);
TCLAP::SwitchArg enableTrackingLossDetection("", "para_enable_tracking_loss", "Enable the tracking loss detection!", cmd, paras.enableTrackingLossDetection);
TCLAP::ValueArg<double> psrThreshold("", "para_psr_threshold",
"if psr is lower than psr threshold, tracking will stop",
false, paras.psrThreshold, "double", cmd);
TCLAP::ValueArg<int> psrPeakDel("", "para_psr_peak_del", "amount of pixels that are deleted"
"for psr calculation around the peak (1 means that a window of 3 by 3 is deleted; 0 means"
"that max response is deleted; 2 * peak_del + 1 pixels are deleted)",
false, paras.psrPeakDel, "integer", cmd);
cmd.parse(argc, argv);
paras.padding = padding.getValue();
paras.outputSigmaFactor = outputSigmaFactor.getValue();
paras.lambda = lambda.getValue();
paras.learningRate = interpFactor.getValue();
paras.cellSize = cellSize.getValue();
paras.scaleSigmaFactor = scaleSigmaFactor.getValue();
paras.scaleStep = scaleStep.getValue();
paras.scaleCellSize = scaleCellSize.getValue();
paras.numberOfScales = numberOfScales.getValue();
paras.psrThreshold = psrThreshold.getValue();
paras.psrPeakDel = psrPeakDel.getValue();
paras.templateSize = templateSize.getValue();
paras.enableTrackingLossDetection = enableTrackingLossDetection.getValue();
// use original paper parameters from
// Danelljan, Martin, et al., "Accurate scale estimation for robust visual tracking," in Proc. BMVC, 2014
if (originalVersion.getValue())
{
paras.padding = static_cast<double>(1);
paras.outputSigmaFactor = static_cast<double>(1.0 / 16.0);
paras.lambda = static_cast<double>(0.01);
paras.learningRate = static_cast<double>(0.025);
paras.templateSize = 100;
paras.cellSize = 1;
paras.enableTrackingLossDetection = false;
paras.psrThreshold = 0;
paras.psrPeakDel = 1;
paras.enableScaleEstimator = true;
paras.scaleSigmaFactor = static_cast<double>(0.25);
paras.scaleStep = static_cast<double>(1.02);
paras.scaleCellSize = 4;
paras.numberOfScales = 33;
paras.originalVersion = true;
paras.resizeType = cv::INTER_AREA;
}
if (debugOutput.getValue())
{
setTrackerDebug(&_debug);
return new cf_tracking::DsstTracker(paras, &_debug);
}
return new cf_tracking::DsstTracker(paras);
}
virtual cf_tracking::CfTracker* parseTrackerParas(TCLAP::CmdLine& cmd, int argc, const char** argv)
{
cf_tracking::KcfParameters paras;
TCLAP::SwitchArg debugOutput("v", "debug", "Output Debug info!", cmd, false);
TCLAP::SwitchArg originalVersion("", "original_version", "Parameters and performance as close to the KCF VOT version as possible.", cmd, false);
TCLAP::SwitchArg originalParametersWithScaleFilter("", "original_parameters_scale_filter", "KCF VOT version parameters with DSST scale filter.", cmd, false);
TCLAP::ValueArg<int> templateSize("", "para_template_size", "template size", false, paras.templateSize, "integer", cmd);
TCLAP::ValueArg<int> cellSize("", "para_cell_size", "cell size of fhog", false, paras.cellSize, "integer", cmd);
TCLAP::ValueArg<double> padding("", "para_padding", "padding around the target", false, paras.padding, "double", cmd);
TCLAP::ValueArg<double> lambda("", "para_lambda", "regularization factor", false, paras.lambda, "double", cmd);
TCLAP::ValueArg<double> outputSigmaFactor("", "para_output_sigma_factor", "spatial bandwidth of the target", false, paras.outputSigmaFactor, "double", cmd);
TCLAP::ValueArg<double> scaleStep("", "para_vot_scale_step", "scale_step", false, paras.votScaleStep, "double", cmd);
TCLAP::ValueArg<double> scaleWeight("", "para_vot_scale_weight", "scale_weight", false, paras.votScaleWeight, "double", cmd);
TCLAP::ValueArg<double> interpFactor("", "para_interpFactor", "interpolation factor for learning", false, paras.interpFactor, "double", cmd);
TCLAP::ValueArg<double> kernelSigma("", "para_kernel_sigma", "sigma for Gaussian kernel", false, paras.kernelSigma, "double", cmd);
TCLAP::ValueArg<double> psrThreshold("", "para_psr_threshold", "if psr is lower than "
"psr threshold, target is assumed to be lost", false, paras.psrThreshold, "double", cmd);
TCLAP::ValueArg<int> psrPeakDel("", "para_psr_peak_del", "amount of pixels that are "
"deleted for psr calculation around the peak (1 means that a window of 3 by 3 is "
"deleted; 0 means that max response is deleted; 2 * peak_del + 1 pixels are deleted)", false, paras.psrPeakDel, "integer", cmd);
TCLAP::SwitchArg useDsstScale("", "para_use_dsst_scale", "Uses the DSST scale filter for scale estimation. "
"Disable for more speed!", cmd, paras.useDsstScaleEstimation);
TCLAP::ValueArg<double> scaleSigmaFactor("", "para_dsst_sigma_factor", "DSST: spatial bandwidth of the target", false, paras.scaleSigmaFactor, "double", cmd);
TCLAP::ValueArg<double> scaleEstimatorStep("", "para_dsst_scale_step", "DSST: scale step", false, paras.scaleEstimatorStep, "double", cmd);
TCLAP::ValueArg<double> scaleLambda("", "para_dsst_lambda", "DSST: regularization for scale estimation", false, paras.scaleLambda, "double", cmd);
TCLAP::ValueArg<int> scaleCellSize("", "para_dsst_cell_size", "DSST: hog cell size for scale estimation", false, paras.scaleCellSize, "integer", cmd);
TCLAP::ValueArg<int> numberOfScales("", "para_dsst_scales", "DSST: number of scales", false, paras.numberOfScales, "integer", cmd);
TCLAP::SwitchArg enableTrackingLossDetection("", "para_enable_tracking_loss", "Enables the tracking loss detection!", cmd, paras.enableTrackingLossDetection);
cmd.parse(argc, argv);
paras.padding = padding.getValue();
paras.lambda = lambda.getValue();
paras.outputSigmaFactor = outputSigmaFactor.getValue();
paras.votScaleStep = scaleStep.getValue();
paras.votScaleWeight = scaleWeight.getValue();
paras.templateSize = templateSize.getValue();
paras.interpFactor = interpFactor.getValue();
paras.kernelSigma = kernelSigma.getValue();
paras.cellSize = cellSize.getValue();
paras.psrThreshold = psrThreshold.getValue();
paras.psrPeakDel = psrPeakDel.getValue();
paras.enableTrackingLossDetection = enableTrackingLossDetection.getValue();
paras.useDsstScaleEstimation = useDsstScale.getValue();
paras.scaleSigmaFactor = scaleSigmaFactor.getValue();
paras.scaleEstimatorStep = scaleEstimatorStep.getValue();
paras.scaleLambda = scaleLambda.getValue();
paras.scaleCellSize = scaleCellSize.getValue();
paras.numberOfScales = numberOfScales.getValue();
if (originalVersion.getValue() || originalParametersWithScaleFilter.getValue())
{
paras.padding = 1.5;
paras.lambda = 0.0001;
paras.outputSigmaFactor = 0.1;
paras.votScaleStep = 1.05;
paras.votScaleWeight = 0.95;
paras.templateSize = 100;
paras.interpFactor = 0.012;
paras.kernelSigma = 0.6;
paras.cellSize = 4;
paras.pixelPadding = 0;
paras.enableTrackingLossDetection = false;
if (originalParametersWithScaleFilter.getValue())
{
paras.useVotScaleEstimation = false;
paras.useDsstScaleEstimation = true;
}
else
{
paras.useVotScaleEstimation = true;
paras.useDsstScaleEstimation = false;
}
paras.useFhogTranspose = false;
}
if (debugOutput.getValue())
{
setTrackerDebug(&_debug);
return new cf_tracking::KcfTracker(paras, &_debug);
}
return new cf_tracking::KcfTracker(paras);
}
