def _calculateInputValuesFromFaceBox(self):
currentFaceBox = self._inputEstimator.faceBox
left, right = currentFaceBox.left, currentFaceBox.right
top, bottom = currentFaceBox.top, currentFaceBox.bottom
if self._inputBoundaries == None:
self._inputBoundaries = Boundary(left, right, top, bottom)
self._faceBoxForInput = currentFaceBox
x, y = self._inputValues[:2]
update = False
if not self._inputBoundaries.isInRanges(x=x):
update = True
if x - self._faceBoxForInput.location[0] > 0:
left, right = x - (right - left), x
else:
left, right = x, x + (right - left)
top = self._faceBoxForInput.top
bottom = self._faceBoxForInput.bottom
if not self._inputBoundaries.isInRanges(y=y):
update = True
if y - self._faceBoxForInput.location[1] > 0:
top, bottom = y - (bottom - top), y
else:
top, bottom = y, y + (bottom - top)
left = self._faceBoxForInput.left
right = self._faceBoxForInput.right
if update:
self._inputBoundaries = Boundary(left, right, top, bottom)
self._faceBoxForInput = FaceBox(int(left), int(top), int(right),
int(bottom))
self._pPoints = self._faceBoxForInput.getProjectionPoints()
return self._inputValues
def _calculateInputValuesFromNose(self):
minX, maxX = self._Landmarks[49, 0], self._Landmarks[53, 0]
minY = (self._Landmarks[1, 1] + self._Landmarks[15, 1]) / 2
maxY = (self._Landmarks[4, 1] + self._Landmarks[12, 1]) / 2
self._inputValues[:2] = self._inputValues[:2] - (minX, minY)
minX, maxX, minY, maxY = 0, maxX - minX, 0, maxY - minY
self._inputBoundaries = Boundary(minX, maxX, minY, maxY)
return self._inputValues
def playMapping():
source = 1 # Experiments_Folder + 'Exp999/Exp999.avi' # 'Exp000/Exp000.avi' # 'Exp001/Exp001.avi'
outputSize = (1920, 1080)
#outputSize = (640, 360)
#outputSize = (640, 480)
#outputSize = (1080, 720)
print(outputSize)
boundary = Boundary(0, outputSize[0], 0, outputSize[1])
mappingStr = 'StaticMappingOn' # 'DynamicMappingOn' #
Mapping = DynamicMapping if mappingStr == 'DynamicMappingOn' else StaticMapping #
mappingFunctions = {
#mappingStr + 'MblNtSSDBox': Mapping(TFMobileNetSSDFaceDetector(squaringFaceBox = True), boundary),
#mappingStr + 'YinsLMarks': Mapping(YinsCNNBasedFacialLandmarkDetector(), boundary),
#mappingStr + 'YinsHPose': Mapping(YinsHeadPoseEstimator(), boundary),
mappingStr + 'MrtcnsGaze': Mapping(MuratcansHeadGazer(), boundary)
}
#mappingFunctions = getDefaultMappingFunctions() (720, 480)
#outputSize = (1280, 720) # recordNWrite
names = [k for k, i in mappingFunctions.items()]
s = ''
for n in names:
s += (n + '_')
s = s[:-1]
expName = 'Exp999'
name = expName+'_'+s
#displayTogetherGivenMappingFunctions(source, mappingFunctions, name, outputSize = outputSize)
recordTogetherGivenMappingFunctions(source, mappingFunctions, expName, name, name, outputSize = outputSize)
def getDemosForThreeExampleMappingFunctions(outputSize = (640, 360)):
boundary = Boundary(0, outputSize[0], 0, outputSize[1])
mappingFunction1 = StaticMapping(TFMobileNetSSDFaceDetector(), boundary)
mappingFunction2 = StaticMapping(YinsCNNBasedFacialLandmarkDetector(), boundary)
mappingFunction3 = StaticMapping(CV2Res10SSCNNHeadPoseEstimator(), boundary)
demo1 = InputEstimationDemoWithMappingFunction(mappingFunction1, 'TFMobileNetSSDFaceDetector')
demo2 = InputEstimationDemoWithMappingFunction(mappingFunction2, 'YinsCNNBasedFacialLandmarkDetector')
demo3 = InputEstimationDemoWithMappingFunction(mappingFunction3, 'CV2Res10SSCNNHeadPoseEstimator')
return [demo1, demo2, demo3]
def _initializeInputCalculator(self):
self._outputDependsAnnotations = False
self._inputBoundaries = Boundary()
if isinstance(self._inputEstimator, FaceDetectorABC):
self._recalculateInputValues = self._calculateInputValuesFromFaceBox
self._inputBoundaries = None
elif isinstance(self._inputEstimator, FacialLandmarkDetectorABC):
self._recalculateInputValues = self._calculateInputValuesFromNose
self._outputDependsAnnotations = True
elif isinstance(self._inputEstimator, MuratcansHeadGazer):
width, height = self._inputEstimator.getGazingFrameDimensions()
self._inputBoundaries = Boundary(0, width, 0, height)
self._recalculateInputValues = self._calculateInputValuesFromHeadGaze
elif isinstance(self._inputEstimator, HeadPoseEstimatorABC):
self._inputBoundaries = Boundary(40, 60, -10, 5)
self._recalculateInputValues = self._calculateInputValuesFromHeadPose
class MappingABC(ABC):
def __init__(self, inputEstimator, outputBoundaries, *args, **kwargs):
self._inputEstimator = inputEstimator
self._initializeInputCalculator()
self._inputValues = numpy.zeros((3, ))
self._outputValues = numpy.zeros((2, ))
self._outputBoundaries = outputBoundaries
super().__init__()
def _calculateInputValuesFromFaceBox(self):
currentFaceBox = self._inputEstimator.faceBox
left, right = currentFaceBox.left, currentFaceBox.right
top, bottom = currentFaceBox.top, currentFaceBox.bottom
if self._inputBoundaries == None:
self._inputBoundaries = Boundary(left, right, top, bottom)
self._faceBoxForInput = currentFaceBox
x, y = self._inputValues[:2]
update = False
if not self._inputBoundaries.isInRanges(x=x):
update = True
if x - self._faceBoxForInput.location[0] > 0:
left, right = x - (right - left), x
else:
left, right = x, x + (right - left)
top = self._faceBoxForInput.top
bottom = self._faceBoxForInput.bottom
if not self._inputBoundaries.isInRanges(y=y):
update = True
if y - self._faceBoxForInput.location[1] > 0:
top, bottom = y - (bottom - top), y
else:
top, bottom = y, y + (bottom - top)
left = self._faceBoxForInput.left
right = self._faceBoxForInput.right
if update:
self._inputBoundaries = Boundary(left, right, top, bottom)
self._faceBoxForInput = FaceBox(int(left), int(top), int(right),
int(bottom))
self._pPoints = self._faceBoxForInput.getProjectionPoints()
return self._inputValues
def _calculateInputValuesFromNose(self):
minX, maxX = self._Landmarks[49, 0], self._Landmarks[53, 0]
minY = (self._Landmarks[1, 1] + self._Landmarks[15, 1]) / 2
maxY = (self._Landmarks[4, 1] + self._Landmarks[12, 1]) / 2
self._inputValues[:2] = self._inputValues[:2] - (minX, minY)
minX, maxX, minY, maxY = 0, maxX - minX, 0, maxY - minY
self._inputBoundaries = Boundary(minX, maxX, minY, maxY)
return self._inputValues
def _calculateInputValuesFromHeadPose(self):
return self._inputValues
def _calculateInputValuesFromHeadGaze(self):
return self._inputValues
def _recalculateInputValues(self):
raise NotImplementedError
def _initializeInputCalculator(self):
self._outputDependsAnnotations = False
self._inputBoundaries = Boundary()
if isinstance(self._inputEstimator, FaceDetectorABC):
self._recalculateInputValues = self._calculateInputValuesFromFaceBox
self._inputBoundaries = None
elif isinstance(self._inputEstimator, FacialLandmarkDetectorABC):
self._recalculateInputValues = self._calculateInputValuesFromNose
self._outputDependsAnnotations = True
elif isinstance(self._inputEstimator, MuratcansHeadGazer):
width, height = self._inputEstimator.getGazingFrameDimensions()
self._inputBoundaries = Boundary(0, width, 0, height)
self._recalculateInputValues = self._calculateInputValuesFromHeadGaze
elif isinstance(self._inputEstimator, HeadPoseEstimatorABC):
self._inputBoundaries = Boundary(40, 60, -10, 5)
self._recalculateInputValues = self._calculateInputValuesFromHeadPose
def _estimateInput(self, frame):
if self._outputDependsAnnotations:
annos = self._inputEstimator.estimateInputValuesWithAnnotations(
frame)
self._inputValues, self._pPoints, self._Landmarks = annos
else:
self._inputValues = self._inputEstimator.estimateInputValues(frame)
return self._inputValues
@abstractmethod
def _calculate(self):
raise NotImplementedError
def _updateOutputValues(self):
self._calculate()
self._outputValues = self._outputBoundaries.keepInside(
self._outputValues)
return self._outputValues
def calculateOutputValues(self, frame):
self._estimateInput(frame)
self._recalculateInputValues()
return self._updateOutputValues()
def calculateOutputValuesWithAnnotations(self, frame):
self._outputDependsAnnotations = True
self.calculateOutputValues(frame)
return self._outputValues, self._inputValues, self._pPoints, self._Landmarks
@property
def inputValues(self):
return self._inputValues
@property
def outputValues(self):
return self._outputValues
def getEstimator(self):
return self._inputEstimator
def getOutputBoundaries(self):
return self._outputBoundaries
def getInputBoundaries(self):
return self._inputBoundaries
def buildMappingFunction(name, estimator, outputSize):
mapping = getFrom(name, EstimatorLists.MappingModule)
boundary = Boundary(0, outputSize[0], 0, outputSize[1])
return mapping(estimator, boundary)