private static int GetThreePointShapeType(HTuple sortedIndex)
{
int index = -1;
HTuple hvRet = new HTuple();
HTuple[] hv_SortedIndex = new HTuple[6];
//1th.Input Value is xiao zhong da
hv_SortedIndex[0] = hv_SortedIndex[0].TupleConcat(0);
hv_SortedIndex[0] = hv_SortedIndex[0].TupleConcat(1);
hv_SortedIndex[0] = hv_SortedIndex[0].TupleConcat(2);
//2th.Input Value is da zhong xiao
hv_SortedIndex[1] = hv_SortedIndex[1].TupleConcat(2);
hv_SortedIndex[1] = hv_SortedIndex[1].TupleConcat(1);
hv_SortedIndex[1] = hv_SortedIndex[1].TupleConcat(0);
//3th.Input Value is zhong da xiao
hv_SortedIndex[2] = hv_SortedIndex[2].TupleConcat(2);
hv_SortedIndex[2] = hv_SortedIndex[2].TupleConcat(0);
hv_SortedIndex[2] = hv_SortedIndex[2].TupleConcat(1);
//4th.Input Value is xiao da zhong
hv_SortedIndex[3] = hv_SortedIndex[3].TupleConcat(0);
hv_SortedIndex[3] = hv_SortedIndex[3].TupleConcat(2);
hv_SortedIndex[3] = hv_SortedIndex[3].TupleConcat(1);
//5th.Input Value is zhong xiao da
hv_SortedIndex[4] = hv_SortedIndex[4].TupleConcat(1);
hv_SortedIndex[4] = hv_SortedIndex[4].TupleConcat(0);
hv_SortedIndex[4] = hv_SortedIndex[4].TupleConcat(2);
//6th.Input Value is da xiao zhong
hv_SortedIndex[5] = hv_SortedIndex[5].TupleConcat(1);
hv_SortedIndex[5] = hv_SortedIndex[5].TupleConcat(2);
hv_SortedIndex[5] = hv_SortedIndex[5].TupleConcat(0);
for (int i = 0; i < 6; i++)
{
index = i;
hvRet = sortedIndex.TupleEqualElem(hv_SortedIndex[i]);
if (hvRet[0] == 1 && hvRet[1] == 1 && hvRet[2] == 1)
{
break;
}
}
return(index);
}
private void train_classifier(HTuple hv_DLClassifierHandle, HTuple hv_FileName,
HTuple hv_NumEpochs, HTuple hv_TrainingImages, HTuple hv_TrainingLabels, HTuple hv_ValidationImages,
HTuple hv_ValidationLabels, HTuple hv_LearningRateStepEveryNthEpoch, HTuple hv_LearningRateStepRatio,
HTuple hv_PlotEveryNthEpoch, HTuple hv_WindowHandle)
{
// Stack for temporary objects
HObject[] OTemp = new HObject[20];
// Local iconic variables
HObject ho_BatchImages = null;
// Local control variables
HTuple hv_TrainingErrors = new HTuple(), hv_ValidationErrors = new HTuple();
HTuple hv_LearningRates = new HTuple(), hv_Epochs = new HTuple();
HTuple hv_LossByIteration = new HTuple(), hv_BatchSize = new HTuple();
HTuple hv_MinValidationError = new HTuple(), hv_NumBatchesInEpoch = new HTuple();
HTuple hv_NumTotalIterations = new HTuple(), hv_PlottedIterations = new HTuple();
HTuple hv_TrainSequence = new HTuple(), hv_SelectPercentageTrainingImages = new HTuple();
HTuple hv_TrainingImagesSelected = new HTuple(), hv_TrainingLabelsSelected = new HTuple();
HTuple hv_Epoch = new HTuple(), hv_Iteration = new HTuple();
HTuple hv_BatchStart = new HTuple(), hv_BatchEnd = new HTuple();
HTuple hv_BatchIndices = new HTuple(), hv_BatchImageFiles = new HTuple();
HTuple hv_BatchLabels = new HTuple(), hv_GenParamName = new HTuple();
HTuple hv_GenParamValue = new HTuple(), hv_DLClassifierTrainResultHandle = new HTuple();
HTuple hv_Loss = new HTuple(), hv_CurrentIteration = new HTuple();
HTuple hv_TrainingDLClassifierResultIDs = new HTuple();
HTuple hv_ValidationDLClassifierResultIDs = new HTuple();
HTuple hv_TrainingTop1Error = new HTuple(), hv_ValidationTop1Error = new HTuple();
HTuple hv_LearningRate = new HTuple();
HTupleVector hvec_TrainingPredictedLabels = new HTupleVector(1);
HTupleVector hvec_TrainingConfidences = new HTupleVector(1);
HTupleVector hvec_ValidationPredictedLabels = new HTupleVector(1);
HTupleVector hvec_ValidationConfidences = new HTupleVector(1);
// Initialize local and output iconic variables
HOperatorSet.GenEmptyObj(out ho_BatchImages);
try
{
//For the plot during training,
//we need to concatenate some intermediate results.
hv_TrainingErrors.Dispose();
hv_TrainingErrors = new HTuple();
hv_ValidationErrors.Dispose();
hv_ValidationErrors = new HTuple();
hv_LearningRates.Dispose();
hv_LearningRates = new HTuple();
hv_Epochs.Dispose();
hv_Epochs = new HTuple();
hv_LossByIteration.Dispose();
hv_LossByIteration = new HTuple();
hv_BatchSize.Dispose();
HOperatorSet.GetDlClassifierParam(hv_DLClassifierHandle, "batch_size", out hv_BatchSize);
hv_MinValidationError.Dispose();
hv_MinValidationError = 1;
//
//Create a tuple that includes all the iterations
//where the plot should be computed (including the last ieration).
hv_NumBatchesInEpoch.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_NumBatchesInEpoch = (((((new HTuple(hv_TrainingImages.TupleLength()
)) / (hv_BatchSize.TupleReal()))).TupleFloor())).TupleInt();
}
hv_NumTotalIterations.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_NumTotalIterations = (hv_NumBatchesInEpoch * hv_NumEpochs) - 1;
}
hv_PlottedIterations.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_PlottedIterations = ((((hv_NumBatchesInEpoch * HTuple.TupleGenSequence(
0, hv_NumEpochs - 1, hv_PlotEveryNthEpoch))).TupleConcat(hv_NumTotalIterations))).TupleRound()
;
}
//
//TrainSequence is used for easier indexing of the training data.
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_TrainSequence.Dispose();
HOperatorSet.TupleGenSequence(0, (new HTuple(hv_TrainingImages.TupleLength()
)) - 1, 1, out hv_TrainSequence);
}
//
//Select a subset of the training data set
//in order to obtain a fast approximation
//of the training error during training (plotting).
hv_SelectPercentageTrainingImages.Dispose();
hv_SelectPercentageTrainingImages = 100;
hv_TrainingImagesSelected.Dispose(); hv_TrainingLabelsSelected.Dispose();
select_percentage_dl_classifier_data(hv_TrainingImages, hv_TrainingLabels,
hv_SelectPercentageTrainingImages, out hv_TrainingImagesSelected, out hv_TrainingLabelsSelected);
//
HTuple end_val25 = hv_NumEpochs - 1;
HTuple step_val25 = 1;
#region
for (hv_Epoch = 0; hv_Epoch.Continue(end_val25, step_val25); hv_Epoch = hv_Epoch.TupleAdd(step_val25))
{
//In order to get randomness in each epoch,
//the training set is shuffled every epoch.
{
HTuple ExpTmpOutVar_0;
tuple_shuffle(hv_TrainSequence, out ExpTmpOutVar_0);
hv_TrainSequence.Dispose();
hv_TrainSequence = ExpTmpOutVar_0;
}
HTuple end_val29 = hv_NumBatchesInEpoch - 1;
HTuple step_val29 = 1;
for (hv_Iteration = 0; hv_Iteration.Continue(end_val29, step_val29); hv_Iteration = hv_Iteration.TupleAdd(step_val29))
{
//Select a batch from the training data set.
hv_BatchStart.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_BatchStart = hv_Iteration * hv_BatchSize;
}
hv_BatchEnd.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_BatchEnd = hv_BatchStart + (hv_BatchSize - 1);
}
hv_BatchIndices.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_BatchIndices = hv_TrainSequence.TupleSelectRange(
hv_BatchStart, hv_BatchEnd);
}
hv_BatchImageFiles.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_BatchImageFiles = hv_TrainingImages.TupleSelect(
hv_BatchIndices);
}
hv_BatchLabels.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_BatchLabels = hv_TrainingLabels.TupleSelect(
hv_BatchIndices);
}
//
//Read the image of the current batch.
ho_BatchImages.Dispose();
HOperatorSet.ReadImage(out ho_BatchImages, hv_BatchImageFiles);
//Augment the images to get a better variety of training images.
hv_GenParamName.Dispose();
hv_GenParamName = "mirror";
hv_GenParamValue.Dispose();
hv_GenParamValue = "rc";
{
HObject ExpTmpOutVar_0;
augment_images(ho_BatchImages, out ExpTmpOutVar_0, hv_GenParamName, hv_GenParamValue);
ho_BatchImages.Dispose();
ho_BatchImages = ExpTmpOutVar_0;
}
//
//Train the network with these images and ground truth labels.
hv_DLClassifierTrainResultHandle.Dispose();
HOperatorSet.TrainDlClassifierBatch(ho_BatchImages, hv_DLClassifierHandle,
hv_BatchLabels, out hv_DLClassifierTrainResultHandle);
//You can access the current value of the loss function,
//which should decrease during the training.
hv_Loss.Dispose();
HOperatorSet.GetDlClassifierTrainResult(hv_DLClassifierTrainResultHandle,
"loss", out hv_Loss);
//Store the loss in a tuple .
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
{
HTuple
ExpTmpLocalVar_LossByIteration = hv_LossByIteration.TupleConcat(
hv_Loss);
hv_LossByIteration.Dispose();
hv_LossByIteration = ExpTmpLocalVar_LossByIteration;
}
}
//
//In regular intervals, we want to evaluate
//how well our classifier performs.
hv_CurrentIteration.Dispose();
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
hv_CurrentIteration = ((hv_Iteration + (hv_NumBatchesInEpoch * hv_Epoch))).TupleInt()
;
}
if ((int)(((hv_CurrentIteration.TupleEqualElem(hv_PlottedIterations))).TupleSum()
) != 0)
{
//Plot the progress regularly.
//Evaluate the current classifier on the training and validation set.
hv_TrainingDLClassifierResultIDs.Dispose(); hvec_TrainingPredictedLabels.Dispose(); hvec_TrainingConfidences.Dispose();
apply_dl_classifier_batchwise(hv_TrainingImagesSelected, hv_DLClassifierHandle,
out hv_TrainingDLClassifierResultIDs, out hvec_TrainingPredictedLabels,
out hvec_TrainingConfidences);
hv_ValidationDLClassifierResultIDs.Dispose(); hvec_ValidationPredictedLabels.Dispose(); hvec_ValidationConfidences.Dispose();
apply_dl_classifier_batchwise(hv_ValidationImages, hv_DLClassifierHandle,
out hv_ValidationDLClassifierResultIDs, out hvec_ValidationPredictedLabels,
out hvec_ValidationConfidences);
//Evaluate the top-1 error on each dataset.
hv_TrainingTop1Error.Dispose();
evaluate_dl_classifier(hv_TrainingLabelsSelected, hv_DLClassifierHandle,
hv_TrainingDLClassifierResultIDs, "top1_error", "global", out hv_TrainingTop1Error);
hv_ValidationTop1Error.Dispose();
evaluate_dl_classifier(hv_ValidationLabels, hv_DLClassifierHandle, hv_ValidationDLClassifierResultIDs,
"top1_error", "global", out hv_ValidationTop1Error);
//Concatenate the values for the plot.
hv_LearningRate.Dispose();
HOperatorSet.GetDlClassifierParam(hv_DLClassifierHandle, "learning_rate",
out hv_LearningRate);
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
{
HTuple
ExpTmpLocalVar_TrainingErrors = hv_TrainingErrors.TupleConcat(
hv_TrainingTop1Error);
hv_TrainingErrors.Dispose();
hv_TrainingErrors = ExpTmpLocalVar_TrainingErrors;
}
}
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
{
HTuple
ExpTmpLocalVar_ValidationErrors = hv_ValidationErrors.TupleConcat(
hv_ValidationTop1Error);
hv_ValidationErrors.Dispose();
hv_ValidationErrors = ExpTmpLocalVar_ValidationErrors;
}
}
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
{
HTuple
ExpTmpLocalVar_LearningRates = hv_LearningRates.TupleConcat(
hv_LearningRate);
hv_LearningRates.Dispose();
hv_LearningRates = ExpTmpLocalVar_LearningRates;
}
}
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
{
HTuple
ExpTmpLocalVar_Epochs = hv_Epochs.TupleConcat(
(hv_PlottedIterations.TupleSelect(new HTuple(hv_Epochs.TupleLength()
))) / (hv_NumBatchesInEpoch.TupleReal()));
hv_Epochs.Dispose();
hv_Epochs = ExpTmpLocalVar_Epochs;
}
}
//Plot validation and error against the epochs in order to
//observe the progress of the training.
plot_dl_classifier_training_progress(hv_TrainingErrors, hv_ValidationErrors, hv_LearningRates, hv_Epochs, hv_NumEpochs, hv_WindowHandle);
//如果上面绘制坐标系到窗体过程报错，搞不定的话就用下面这句代替，直接显示训练数据到窗体
//disp_training_progress_datas(hv_TrainingErrors, hv_ValidationErrors, hv_LearningRates, hv_Epochs, hv_NumEpochs, hv_WindowHandle);
if ((int)(new HTuple(hv_ValidationTop1Error.TupleLessEqual(hv_MinValidationError))) != 0)
{
HOperatorSet.WriteDlClassifier(hv_DLClassifierHandle, hv_FileName);
hv_MinValidationError.Dispose();
hv_MinValidationError = new HTuple(hv_ValidationTop1Error);
}
}
}
//Reduce the learning rate every nth epoch.
if ((int)(new HTuple((((hv_Epoch + 1) % hv_LearningRateStepEveryNthEpoch)).TupleEqual(
0))) != 0)
{
using (HDevDisposeHelper dh = new HDevDisposeHelper())
{
HOperatorSet.SetDlClassifierParam(hv_DLClassifierHandle, "learning_rate",
hv_LearningRate * hv_LearningRateStepRatio);
}
hv_LearningRate.Dispose();
HOperatorSet.GetDlClassifierParam(hv_DLClassifierHandle, "learning_rate",
out hv_LearningRate);
}
}
#endregion
// stop(...); only in hdevelop
ho_BatchImages.Dispose();
hv_TrainingErrors.Dispose();
hv_ValidationErrors.Dispose();
hv_LearningRates.Dispose();
hv_Epochs.Dispose();
hv_LossByIteration.Dispose();
hv_BatchSize.Dispose();
hv_MinValidationError.Dispose();
hv_NumBatchesInEpoch.Dispose();
hv_NumTotalIterations.Dispose();
hv_PlottedIterations.Dispose();
hv_TrainSequence.Dispose();
hv_SelectPercentageTrainingImages.Dispose();
hv_TrainingImagesSelected.Dispose();
hv_TrainingLabelsSelected.Dispose();
hv_Epoch.Dispose();
hv_Iteration.Dispose();
hv_BatchStart.Dispose();
hv_BatchEnd.Dispose();
hv_BatchIndices.Dispose();
hv_BatchImageFiles.Dispose();
hv_BatchLabels.Dispose();
hv_GenParamName.Dispose();
hv_GenParamValue.Dispose();
hv_DLClassifierTrainResultHandle.Dispose();
hv_Loss.Dispose();
hv_CurrentIteration.Dispose();
hv_TrainingDLClassifierResultIDs.Dispose();
hv_ValidationDLClassifierResultIDs.Dispose();
hv_TrainingTop1Error.Dispose();
hv_ValidationTop1Error.Dispose();
hv_LearningRate.Dispose();
hvec_TrainingPredictedLabels.Dispose();
hvec_TrainingConfidences.Dispose();
hvec_ValidationPredictedLabels.Dispose();
hvec_ValidationConfidences.Dispose();
return;
}
catch (HalconException HDevExpDefaultException)
{
ho_BatchImages.Dispose();
hv_TrainingErrors.Dispose();
hv_ValidationErrors.Dispose();
hv_LearningRates.Dispose();
hv_Epochs.Dispose();
hv_LossByIteration.Dispose();
hv_BatchSize.Dispose();
hv_MinValidationError.Dispose();
hv_NumBatchesInEpoch.Dispose();
hv_NumTotalIterations.Dispose();
hv_PlottedIterations.Dispose();
hv_TrainSequence.Dispose();
hv_SelectPercentageTrainingImages.Dispose();
hv_TrainingImagesSelected.Dispose();
hv_TrainingLabelsSelected.Dispose();
hv_Epoch.Dispose();
hv_Iteration.Dispose();
hv_BatchStart.Dispose();
hv_BatchEnd.Dispose();
hv_BatchIndices.Dispose();
hv_BatchImageFiles.Dispose();
hv_BatchLabels.Dispose();
hv_GenParamName.Dispose();
hv_GenParamValue.Dispose();
hv_DLClassifierTrainResultHandle.Dispose();
hv_Loss.Dispose();
hv_CurrentIteration.Dispose();
hv_TrainingDLClassifierResultIDs.Dispose();
hv_ValidationDLClassifierResultIDs.Dispose();
hv_TrainingTop1Error.Dispose();
hv_ValidationTop1Error.Dispose();
hv_LearningRate.Dispose();
hvec_TrainingPredictedLabels.Dispose();
hvec_TrainingConfidences.Dispose();
hvec_ValidationPredictedLabels.Dispose();
hvec_ValidationConfidences.Dispose();
throw HDevExpDefaultException;
}
}