void LHENode<ImageSigT>::clheProcess(const Matrix<float>& img,Matrix<float>& result)
{
if (m_img_update_time < getOutputPort()->getMTime())
{
buildPDFGrid(img);
clheBuildCDFGrid();
m_img_update_time = getOutputPort()->getMTime();
}
sliceCDFGrid(img,result);
}
int FilterTypeInfoHelper::portType(const QString &id) const
{
const PortInfo& input = getInputPort(id);
if(input.isValid())
return input.type;
const PortInfo& output = getOutputPort(id);
if(output.isValid())
return output.type;
return -1;
}
void DenseDescriptorDictionaryTrainer::train()
{
//parse training images
parseImage();
//write to output signal
InfoSignal<std::string>* output_signal_info = TO_INFO_SIGNAL(std::string,getOutputPort(OUTPUT_PORT_DIC_FILE_INFO));
m_trainer_info = m_trainer_name;
output_signal_info->info = &m_trainer_info;
EAGLEEYE_INFO("success to build dictionary (%s)\n",m_trainer_name.c_str());
}
void NoximReservationTable::reserve(const int port_in, const int port_out) {
// reservation of reserved/not valid ports is illegal. Correctness
// should be assured by NoximReservationTable users
assert(isAvailable(port_out));
// check for previous reservation to be released
int port = getOutputPort(port_in);
if (port != NOT_RESERVED)
release(port);
rtable[port_out] = port_in;
}
GUI::GUI() {
IGTLinkStreamer::pointer streamer = IGTLinkStreamer::New();
setWidth(getScreenWidth());
setHeight(getScreenHeight());
enableMaximized();
setTitle("FAST Neural Network Despeckling");
auto resizer = ImageResizer::New();
resizer->setInputConnection(streamer->getOutputPort());
resizer->setWidth(512);
resizer->setHeight(512);
UltrasoundImageEnhancement::pointer enhancer1 = UltrasoundImageEnhancement::New();
enhancer1->setInputConnection(resizer->getOutputPort());
ImageRenderer::pointer renderer1 = ImageRenderer::New();
renderer1->addInputConnection(enhancer1->getOutputPort());
View* viewOrig = createView();
viewOrig->setBackgroundColor(Color::Black());
viewOrig->set2DMode();
viewOrig->addRenderer(renderer1);
ImageToImageNetwork::pointer network = ImageToImageNetwork::New();
network->load("/home/smistad/Downloads/filname_test.pb");
network->addOutputNode(0, "activation_8/Tanh:0", NodeType::TENSOR);
network->setSignedInputNormalization(true);
network->setScaleFactor(1.0f/255.0f);
network->setInputConnection(streamer->getOutputPort());
UltrasoundImageEnhancement::pointer enhancer2 = UltrasoundImageEnhancement::New();
enhancer2->setInputConnection(network->getOutputPort());
ImageRenderer::pointer renderer2 = ImageRenderer::New();
renderer2->addInputConnection(enhancer2->getOutputPort());
View* view = createView();
view->setBackgroundColor(Color::Black());
view->set2DMode();
view->addRenderer(renderer2);
QHBoxLayout* layout = new QHBoxLayout;
layout->addWidget(viewOrig);
layout->addWidget(view);
mWidget->setLayout(layout);
}
bool DenseDescriptorDictionaryTrainer::isNeedProcessed()
{
if(EagleeyeIO::isexist(m_trainer_folder + m_trainer_name + m_ext_name))
{
//dictionary has been existed
EAGLEEYE_INFO("dictionary has been existed \n");
//write output signal info directly
InfoSignal<std::string>* output_signal_info = TO_INFO_SIGNAL(std::string,getOutputPort(OUTPUT_PORT_DIC_FILE_INFO));
m_trainer_info = m_trainer_name;
output_signal_info->info = &m_trainer_info;
return false;
}
else
{
return true;
}
}
bool SemanticBagOfWordsTrainer::isNeedProcessed()
{
if (EagleeyeIO::isexist(m_trainer_folder + m_trainer_name + m_ext_name))
{
//check whether classification parameters file has been built
EAGLEEYE_INFO("classifier model have existed \n");
//write classifier model info
InfoSignal<std::string>* output_signal_calssifier_info =
TO_INFO_SIGNAL(std::string,getOutputPort(OUTPUT_PORT_CLASSIFIER_INFO));
m_trainer_info = m_trainer_name;
output_signal_calssifier_info->info = &m_trainer_info;
return false;
}
else
return true;
}
bool FilterTypeInfoHelper::hasOutputPort(const QString &id) const
{
return getOutputPort(id).isValid();
}
void SemanticBagOfWordsTrainer::train()
{
//get samples file
std::string training_samples_file = TO_INFO(std::string,getInputPort(INPUT_PORT_SAMPLES_INFO));
Matrix<float> samples_label,samples_representation;
{
Matrix<float> training_samples;
EagleeyeIO::read(training_samples,m_trainer_folder + training_samples_file + m_ext_name,READ_BINARY_MODE);
int samples_num = training_samples.rows();
samples_label = training_samples(Range(0,samples_num),Range(0,1));
samples_label.clone();
samples_representation = training_samples(Range(0,samples_num),Range(1,training_samples.cols()));
samples_representation.clone();
}
EagleeyeIO semantic_models_o;
semantic_models_o.createWriteHandle(m_trainer_folder + m_trainer_name + m_ext_name,false,WRITE_BINARY_MODE);
//write class number
semantic_models_o.write(m_class_num);
for (int class_index = 0; class_index < 2; ++class_index)
{
EAGLEEYE_INFO("process model %d\n",class_index);
Matrix<float> resampling_label = samples_label;
resampling_label.clone();
int samples_num = resampling_label.rows();
//reassign positive and negative samples
for (int i = 0; i < samples_num; ++i)
{
if (int(resampling_label(i)) == class_index)
resampling_label(i) = 0.0f;
else
resampling_label(i) = 1.0f;
}
Matrix<float> resampling_samples = samples_representation;
resampling_samples.clone();
resampling(resampling_samples,resampling_label);
int resamples_num = resampling_label.rows();
//reconstruct sample feature
Matrix<float> words_frequency = resampling_samples(Range(0,resamples_num),Range(0,m_words_num));
Matrix<float> words_pair_dis = resampling_samples(Range(0,resamples_num),Range(m_words_num,m_words_num + m_clique_num));
Matrix<float> words_pair_angle = resampling_samples(Range(0,resamples_num),Range(m_words_num + m_clique_num,m_words_num + 2 * m_clique_num));
Matrix<int> target_states(resamples_num,m_words_num);
for (int sample_index = 0; sample_index < resamples_num; ++sample_index)
{
if (int(resampling_label(sample_index)) == 0)
{
//positive samples
for (int w_index = 0; w_index < m_words_num; ++w_index)
{
if (words_frequency(sample_index,w_index) > 0.0001f)
target_states(sample_index,w_index) = 1;
else
target_states(sample_index,w_index) = 0;
}
}
else
{
//negative samples
for (int w_index = 0; w_index < m_words_num; ++w_index)
{
if (words_frequency(sample_index,w_index) > 0.0001f)
target_states(sample_index,w_index) = 2;
else
target_states(sample_index,w_index) = 0;
}
}
}
/* putToMatlab(target_states,"target");*/
//start training
SemanticBagOfWords* semantic_model = new SemanticBagOfWords(m_words_num,3);
semantic_model->semanticLearn(target_states,words_frequency,words_pair_dis,words_pair_angle);
Matrix<float> unary_coe,clique_coe;
semantic_model->getModelParam(unary_coe,clique_coe);
semantic_models_o.write(unary_coe);
semantic_models_o.write(clique_coe);
delete semantic_model;
}
semantic_models_o.destroyHandle();
//write classifier model info
InfoSignal<std::string>* output_signal_calssifier_info =
TO_INFO_SIGNAL(std::string,getOutputPort(OUTPUT_PORT_CLASSIFIER_INFO));
m_trainer_info = m_trainer_name;
output_signal_calssifier_info->info = &m_trainer_info;
//evaluate performance
}
Matrix<typename ImgSigT::MetaType> ImageIONode<ImgSigT>::getImage()
{
return TO_IMAGE(PixelType,getOutputPort(0));
}