public frmNnClassifier(NnClassifier classifier, frmMain frmMain)
{
InitializeComponent();
_frmMain = frmMain;
_classifier = classifier;
FillWithConfiguration();
}
public void RejectPatches()
{
// arrange - create object model
ObjectModelStub objectModel = new ObjectModelStub();
// arrange - create scanning window generator with dummy windows
ScanningWindowGeneratorStub swg = new ScanningWindowGeneratorStub();
swg.ScanningWindows = new IBoundingBox[]
{
new BoundingBox(new PointF(10, 10), new SizeF(5, 5)),
new BoundingBox(new PointF(10, 10), new SizeF(5, 5)),
new BoundingBox(new PointF(10, 10), new SizeF(5, 5)),
};
// arrange - create nn classifier
NnClassifier classifier = new NnClassifier(swg, objectModel, 0.75f, 0.6f);
classifier.PostInstantiation();
classifier.Initialize(new Image <Gray, byte>(new Size(0, 0)), new BoundingBox(new PointF(), new SizeF()));
// arrange - hard code relative similarities of patches in scanning windows
// 1. relative similarity of the patch in the first scanning window
// is below the threshold -> scanning window is rejected
// 2. relative similarity of the patch in the second scanning window
// is the same as the threshold -> scanning window is rejected
// 3. relative similarity of the patch in the third scanning window
// is above the threshold -> scanning window is accepted
objectModel.RelativeSimilarities = new List <float>()
{
0.1f, 0.6f, 0.9f
};
objectModel.PnnSimilarities = new List <float>()
{
0, 0.9f, 0.9f
};
objectModel.NnnSimilarities = new List <float>()
{
1, 0.9f, 0
};
// define expected accepted scanning windows
List <int> expected = new List <int>()
{
2
};
// get actual accepted scanning windows
List <int> actual = classifier.AcceptedWindows(new Image <Gray, byte>(new Size(1, 1)), new List <int>()
{
0, 1, 2
});
// assert
CollectionAssert.AreEqual(expected, actual);
}
public void RejectPatches()
{
// arrange - create object model
IObjectModel objectModel = new ObjectModel(new Size(15, 15));
objectModel.PostInstantiation();
// -> create 2 positive and 2 negative patches
Image <Gray, byte> helpImage = new Image <Gray, byte>(Path.Combine(_resourceDir, "violeta_1.jpg"));
Image <Gray, byte> posPatch1 = helpImage.GetPatch(new PointF(168, 178), new Size(80, 40)).Resize(objectModel.PatchSize.Width, objectModel.PatchSize.Height, INTER.CV_INTER_LINEAR);
Image <Gray, byte> posPatch2 = helpImage.GetPatch(new PointF(169, 179), new Size(81, 41)).Resize(objectModel.PatchSize.Width, objectModel.PatchSize.Height, INTER.CV_INTER_LINEAR);
Image <Gray, byte> negPatch1 = helpImage.GetPatch(new PointF(300, 200), new Size(50, 50)).Resize(objectModel.PatchSize.Width, objectModel.PatchSize.Height, INTER.CV_INTER_LINEAR);
Image <Gray, byte> negPatch2 = helpImage.GetPatch(new PointF(200, 300), new Size(50, 50)).Resize(objectModel.PatchSize.Width, objectModel.PatchSize.Height, INTER.CV_INTER_LINEAR);
// -> add patches to the object model
objectModel.AddPositivePatch(posPatch1);
objectModel.AddPositivePatch(posPatch2);
objectModel.AddNegativePatch(negPatch1);
objectModel.AddNegativePatch(negPatch2);
// arrange - create scanning window generator with stub windows
ScanningWindowGeneratorStub swg = new ScanningWindowGeneratorStub();
swg.ScanningWindows = new IBoundingBox[]
{
new BoundingBox(new PointF(167, 177), new SizeF(79, 39)),
new BoundingBox(new PointF(300, 300), new SizeF(80, 40))
};
// arrange - create classifier
NnClassifier classifier = new NnClassifier(swg, objectModel, 0.75f, 0.6f);
classifier.PostInstantiation();
classifier.Initialize(new Image <Gray, byte>(new Size(0, 0)), new BoundingBox(new PointF(), new SizeF()));
// define expected accepted patches
List <int> expected = new List <int>()
{
0
};
// get actual accepted patches
List <int> actual = classifier.AcceptedWindows(helpImage, new List <int>()
{
0, 1
});
// assert
CollectionAssert.AreEqual(expected, actual);
}
public void RejectPatches_FrameRoiIsNotSetAfterProcessing()
{
// arrange - create object model
IObjectModel objectModel = new ObjectModel(new Size(15, 15));
objectModel.PostInstantiation();
// -> create 1 patch
Image <Gray, byte> helpImage = new Image <Gray, byte>(Path.Combine(_resourceDir, "violeta_1.jpg"));
Image <Gray, byte> patch = helpImage.GetPatch(new PointF(168, 178), new Size(80, 40)).Resize(objectModel.PatchSize.Width, objectModel.PatchSize.Height, INTER.CV_INTER_LINEAR);
// -> add patch to the object model
objectModel.AddPositivePatch(patch);
// arrange - create scanning window generator with stub windows
ScanningWindowGeneratorStub swg = new ScanningWindowGeneratorStub();
swg.ScanningWindows = new IBoundingBox[]
{
new BoundingBox(new PointF(167, 177), new SizeF(79, 39))
};
// arrange - create classifier
NnClassifier classifier = new NnClassifier(swg, objectModel, 0.75f, 0.6f);
classifier.PostInstantiation();
classifier.Initialize(new Image <Gray, byte>(new Size(0, 0)), new BoundingBox(new PointF(), new SizeF()));
// define expected frame ROI after processing
Rectangle expected = new Rectangle(new Point(), helpImage.Size);
// get actual frame ROI after processing
classifier.AcceptedWindows(helpImage, new List <int>()
{
0
});
Rectangle actual = helpImage.ROI;
// assert
Assert.AreEqual(expected, actual);
Assert.IsTrue(!helpImage.IsROISet); // just in case
}
private static IDetector LoadDetector(IObjectModel objectModel)
{
// scanning window generator
IScanningWindowGenerator scanningWindowGenerator = new ScanningWindowGenerator
(
1.2f,
0.1f,
0.1f,
100
);
// cascaded classifier
// variance classifier
IClassifier varianceClassifier = new VarianceClassifier(scanningWindowGenerator, 0.5);
// ensemble classifier
int width = objectModel.PatchSize.Width;
int height = objectModel.PatchSize.Height;
IPixelComparisonScheduler pcs = new PixelComparisonScheduler(12, 3, width, height);
IClassifier ensembleClassifier = new EnsembleClassifier(scanningWindowGenerator, pcs, 2);
// nn classifier
NnClassifier nnClassifier = new NnClassifier(scanningWindowGenerator, objectModel, 0.75f, 0.55f);
// instantiate cascaded classifier
IClassifier cascadedClassifier = new CascadedClassifier(varianceClassifier, ensembleClassifier, nnClassifier);
IDetector detector = new Detector
(
scanningWindowGenerator,
cascadedClassifier,
Service.NonMaximalBoundingBoxSuppress
);
return(detector);
}
public Learner(
IObjectModel objectModel,
IDetector detector,
PositivePatchSynthesisInfo initPositivePatchSysthesisInfo,
NegativePatchPickingInfo initNegativePatchPickingInfo,
PositivePatchSynthesisInfo runtimePositivePatchSysthesisInfo,
NegativePatchPickingInfo runtimeNegativePatchPickingInfo,
float sameSimilarityThreshold,
float validConservativeSimilarityThreshold
)
{
_objectModel = objectModel;
_detector = detector;
_initPositivePatchSynthesisInfo = initPositivePatchSysthesisInfo;
_initNegativePatchPickingInfo = initNegativePatchPickingInfo;
_runtimePosPatchSynthesisInfo = runtimePositivePatchSysthesisInfo;
_runtimeNegativePatchPickingInfo = runtimeNegativePatchPickingInfo;
_sameSimilarityThreshold = sameSimilarityThreshold;
_validConservativeSimilarityThreshold = validConservativeSimilarityThreshold;
_varianceClassifier = ((_detector as Detector).CascadedClassifier as CascadedClassifier).VarianceClassifier as VarianceClassifier;
_ensembleClassifier = ((_detector as Detector).CascadedClassifier as CascadedClassifier).EnsembleClassifier as EnsembleClassifier;
_nnClassifier = ((_detector as Detector).CascadedClassifier as CascadedClassifier).NnClassifier as NnClassifier;
}