int main(int argc, char **argv) {
// InitVars is part of base. We need because of the vars in dp_payoffs.cpp
InitVars(argc, argv);
// Declare the supported options.
po::options_description desc("Allowed options");
desc.add_options()
("help", "produce help message")
("features", po::value<string>()->required(), "file containing payoff features")
("weights", po::value<string>(), "feature weights")
("corner_penalty", po::value<float>()->required(), "per-corner penalty")
("occlusion_penalty", po::value<float>()->required(),
"additional penalty for occluding corners")
("with_gradient", "also compute gradient")
("logit", "use logistic likelihood (default is Gaussian)")
("delta", po::value<double>()->default_value(kDefaultDelta),
"Window size for finite differences.")
;
// Parse options
po::variables_map opts;
try {
po::store(po::parse_command_line(argc, argv, desc), opts);
po::notify(opts);
} catch (const po::required_option& ex) {
cout << "Missing required option: " << ex.get_option_name() << "\n" << desc << "\n";
return -1;
}
if (opts.count("help")) {
cout << desc << "\n";
return -1;
}
if (opts.count("with_gradient")) {
cout << "GRADIENTS NOT IMPLEMENTED";
return -1;
}
// Read command line arguments
ManhattanHyperParameters params;
params.corner_penalty = opts["corner_penalty"].as<float>();
params.occlusion_penalty = opts["occlusion_penalty"].as<float>();
params.weights = stream_to<VecF>(opts["weights"].as<string>());
// Evaluate the likelihood
double loglik = EvaluateLikelihood(params,
opts["features"].as<string>(),
opts.count("logit_likelihood")>0);
// Done. Print with high precision so matlab can read with high precision.
DLOG << format("%.18e ") % loglik;
return 0;
}
CRgbImage CSkinFeatureExtractor::SkinDetector(const IRgbImage &img) const
{
CRgbImage *InImage = (CRgbImage*)&img;
int width = InImage->Width();
int height = InImage->Height();
// allcoate memory
DOUBLE **ppLikelihoodOnSkin = new double*[clusternum];
DOUBLE *pLikelihoodOffSkin;
for (int i = 0; i < clusternum; ++i)
{
ppLikelihoodOnSkin[i] = new double[width * height];
}
pLikelihoodOffSkin = new double[width * height];
// detect each pixel to obtain likelihood
for (int y = 0; y < height; ++y)
{
for (int x = 0; x < width; ++x)
{
const BYTE R = (*InImage)(x, y).r;
const BYTE G = (*InImage)(x, y).g;
const BYTE B = (*InImage)(x, y).b;
//const BYTE A = img(x, y).A();
DOUBLE *pLikelihoodOnSkin = new double[clusternum];
DOUBLE dLikelihoodOffSkin;
EvaluateLikelihood(R, G, B, pLikelihoodOnSkin, dLikelihoodOffSkin);
DWORD index = y * width + x;
for (int i = 0; i < clusternum; ++i)
{
ppLikelihoodOnSkin[i][index] = pLikelihoodOnSkin[i];
}
pLikelihoodOffSkin[index] = dLikelihoodOffSkin;
delete []pLikelihoodOnSkin;
}
}
// process
CRgbImage m_SoftMap;
//m_SoftMap.Allocate(InImage);
//m_SoftMap = &InImage.Clone();
m_SoftMap.Copy(*InImage);
int iSkinType = 0;
SkinAdapter((const DOUBLE**)ppLikelihoodOnSkin, pLikelihoodOffSkin, m_SoftMap, iSkinType);
//m_SoftMap.Save(L"temp.jpg");
// skin segmentation
//CRgbImage m_SegmentColorMap;
//SkinSegmentor(ppLikelihoodOnSkin[iSkinType], pLikelihoodOffSkin, InImage, m_SegmentColorMap);
// free memory
for (int i = 0; i < clusternum; ++i)
{
delete []ppLikelihoodOnSkin[i];
}
delete []ppLikelihoodOnSkin;
delete []pLikelihoodOffSkin;
/*
if (m_bEnableFaceDetection)
{
// face detection
CImageRgb rgbImage;
rgbImage.Copy(InImage);
FaceDetector(rgbImage);
}
if (m_bEnableROIDetection)
{
// ROI detection
int iFocusLeft = 0;
int iFocusTop = 0;
int iWidth = 0;
int iHeight = 0;
ROIDetector(InImage, iFocusLeft, iFocusTop, iWidth, iHeight);
// create detection result with ROI and face detection
CreateROIImage(m_SegmentColorMap, iFocusLeft, iFocusTop, iWidth, iHeight);
}
*/
return m_SoftMap;
}
