private void DoCompletion(object sender, DoWorkEventArgs e)
{
ShapeModelInference inference = new ShapeModelInference();
inference.SetBeliefs(this.shapeModel);
inference.MaskCompletionIterationCount = 300;
inference.IterationsBetweenCallbacks = 5;
inference.MaskCompletionProgress += OnMaskCompletionProgressChange;
Mask completedMask = inference.CompleteMasks(new[] { this.maskToComplete }, new[] { (bool[,])e.Argument })[0];
e.Result = GetCertainCompletedMask(completedMask);
}
private static void Main()
{
Rand.Restart(666);
//AngleScaleTest();
//return;
const int shapePartCount = 8;
const int traitCount = 5;
const double observationNoiseProb = 0.01;
const int trainingSetSize = 30;
const int imageRepeatCount = 1;
const string modelNamePrefx = "horses";
//const string trainTestImagePath = "../../../Data/Caltech101/AnnotationsConverted/Motorbikes_16";
//const string trainTestImagePath = "../../../Data/horses/scale_invariance";
const string trainTestImagePath = "../../../Data/horses/clean";
bool[][,] trainingLabels = ImageHelpers.LoadImagesAsSameSizeMasks(trainTestImagePath, trainingSetSize, trainingSetSize, 0);
trainingLabels = RepeatImages(trainingLabels, imageRepeatCount);
// Save training set
for (int i = 0; i < trainingLabels.Length; ++i)
{
ImageHelpers.ArrayToBitmap(trainingLabels[i], b => b ? Color.Red : Color.Green).Save(string.Format("training_labels_{0}.png", i));
}
ShapeModelInference inference = new ShapeModelInference();
inference.TrainingIterationCount = 10000;
inference.IterationsBetweenCallbacks = 1;
inference.SetPriorBeliefs(trainingLabels[0].GetLength(0), trainingLabels[0].GetLength(1), traitCount, shapePartCount, observationNoiseProb);
inference.ShapeModelLearningProgress += OnShapeModelLearningProgress;
ShapeModel model = inference.LearnModel(trainingLabels).Item1;
model.Save(string.Format("./{0}_{1}_traits_{2}_images_{3}_parts.bin", modelNamePrefx, traitCount, trainingSetSize, shapePartCount));
}
private void DoFitting(object sender, DoWorkEventArgs e)
{
ShapeModelInference inference = new ShapeModelInference();
inference.SetBeliefs(this.shapeModel);
inference.TrainingIterationCount = 1000;
inference.IterationsBetweenCallbacks = 5;
inference.ShapeModelLearningProgress += OnShapeFittingProgressChange;
e.Result = inference.LearnModel(new[] { this.maskToComplete }, false);
}