class Main(QMainWindow):
haveData = pyqtSignal()
dataReadyToView = pyqtSignal()
def __init__(self, argv):
QMainWindow.__init__(self)
#Normalize the data if true
self._normalize_data=True
if 'notnormalize' in sys.argv:
print sys.argv
self._normalize_data=False
sys.argv.remove('notnormalize')
self._colorTable16 = self._createDefault16ColorColorTable()
self.g = Graph()
self.workflow = None
self.fixableOperators = []
self.featureDlg=None
#The old ilastik provided the following scale names:
#['Tiny', 'Small', 'Medium', 'Large', 'Huge', 'Megahuge', 'Gigahuge']
#The corresponding scales are:
self.featScalesList=[0.3, 0.7, 1, 1.6, 3.5, 5.0, 10.0]
self.initUic()
#if the filename was specified on command line, load it
def loadFile():
self._openFile(sys.argv[1:])
if len(sys.argv) > 2:
QTimer.singleShot(0, loadFile)
def setIconToViewMenu(self):
self.actionOnly_for_current_view.setIcon(QIcon(self.editor.imageViews[self.editor._lastImageViewFocus]._hud.axisLabel.pixmap()))
def initUic(self):
p = os.path.split(__file__)[0]+'/'
if p == "/": p = "."+p
uic.loadUi(p+"/classificationWorkflow.ui", self)
#connect the window and graph creation to the opening of the file
self.actionOpenFile.triggered.connect(self.openFile)
self.actionOpenStack.triggered.connect(self.openImageStack)
self.actionQuit.triggered.connect(qApp.quit)
def toggleDebugPatches(show):
self.editor.showDebugPatches = show
def fitToScreen():
shape = self.editor.posModel.shape
for i, v in enumerate(self.editor.imageViews):
s = list(copy.copy(shape))
del s[i]
v.changeViewPort(v.scene().data2scene.mapRect(QRectF(0,0,*s)))
def fitImage():
if hasattr(self.editor, '_lastImageViewFocus'):
self.editor.imageViews[self.editor._lastImageViewFocus].fitImage()
def restoreImageToOriginalSize():
if hasattr(self.editor, '_lastImageViewFocus'):
self.editor.imageViews[self.editor._lastImageViewFocus].doScaleTo()
def rubberBandZoom():
if hasattr(self.editor, '_lastImageViewFocus'):
if not self.editor.imageViews[self.editor._lastImageViewFocus]._isRubberBandZoom:
self.editor.imageViews[self.editor._lastImageViewFocus]._isRubberBandZoom = True
self.editor.imageViews[self.editor._lastImageViewFocus]._cursorBackup = self.editor.imageViews[self.editor._lastImageViewFocus].cursor()
self.editor.imageViews[self.editor._lastImageViewFocus].setCursor(Qt.CrossCursor)
else:
self.editor.imageViews[self.editor._lastImageViewFocus]._isRubberBandZoom = False
self.editor.imageViews[self.editor._lastImageViewFocus].setCursor(self.editor.imageViews[self.editor._lastImageViewFocus]._cursorBackup)
def hideHud():
if self.editor.imageViews[0]._hud.isVisible():
hide = False
else:
hide = True
for i, v in enumerate(self.editor.imageViews):
v.hideHud(hide)
def toggleSelectedHud():
if hasattr(self.editor, '_lastImageViewFocus'):
self.editor.imageViews[self.editor._lastImageViewFocus].toggleHud()
def centerAllImages():
for i, v in enumerate(self.editor.imageViews):
v.centerImage()
def centerImage():
if hasattr(self.editor, '_lastImageViewFocus'):
self.editor.imageViews[self.editor._lastImageViewFocus].centerImage()
self.actionOnly_for_current_view.setEnabled(True)
self.actionCenterAllImages.triggered.connect(centerAllImages)
self.actionCenterImage.triggered.connect(centerImage)
self.actionToggleAllHuds.triggered.connect(hideHud)
self.actionToggleSelectedHud.triggered.connect(toggleSelectedHud)
self.actionShowDebugPatches.toggled.connect(toggleDebugPatches)
self.actionFitToScreen.triggered.connect(fitToScreen)
self.actionFitImage.triggered.connect(fitImage)
self.actionReset_zoom.triggered.connect(restoreImageToOriginalSize)
self.actionRubberBandZoom.triggered.connect(rubberBandZoom)
self.haveData.connect(self.initGraph)
self.dataReadyToView.connect(self.initEditor)
self.layerstack = LayerStackModel()
model = LabelListModel()
self.labelListView.setModel(model)
self.labelListModel=model
self.labelListModel.rowsAboutToBeRemoved.connect(self.onLabelAboutToBeRemoved)
self.labelListModel.labelSelected.connect(self.switchLabel)
def onDataChanged(topLeft, bottomRight):
firstRow = topLeft.row()
lastRow = bottomRight.row()
firstCol = topLeft.column()
lastCol = bottomRight.column()
if lastCol == firstCol == 0:
assert(firstRow == lastRow) #only one data item changes at a time
#in this case, the actual data (for example color) has changed
self.switchColor(firstRow+1, self.labelListModel[firstRow].color)
self.editor.scheduleSlicesRedraw()
else:
#this column is used for the 'delete' buttons, we don't care
#about data changed here
pass
self.labelListModel.dataChanged.connect(onDataChanged)
self.AddLabelButton.clicked.connect(self.addLabel)
self.SelectFeaturesButton.clicked.connect(self.onFeatureButtonClicked)
self.checkInteractive.setEnabled(False)
self.checkInteractive.toggled.connect(self.toggleInteractive)
self.interactionComboBox.currentIndexChanged.connect(self.changeInteractionMode)
self.interactionComboBox.setEnabled(False)
self._initFeatureDlg()
def toggleInteractive(self, checked):
print "toggling interactive mode to '%r'" % checked
#Check if the number of labels in the layer stack is equals to the number of Painted labels
if checked==True:
labels =numpy.unique(numpy.asarray(self.workflow.labels.outputs["nonzeroValues"][:].allocate().wait()[0]))
nPaintedLabels=labels.shape[0]
nLabelsLayers = self.labelListModel.rowCount()
selectedFeatures = numpy.asarray(self.featureDlg.featureTableWidget.createSelectedFeaturesBoolMatrix())
if nPaintedLabels!=nLabelsLayers:
self.checkInteractive.setCheckState(0)
mexBox=QMessageBox()
mexBox.setText("Did you forget to paint some labels?")
mexBox.setInformativeText("Painted Labels %d \nNumber Active Labels Layers %d"%(nPaintedLabels,self.labelListModel.rowCount()))
mexBox.exec_()
return
if (selectedFeatures==0).all():
self.checkInteractive.setCheckState(0)
mexBox=QMessageBox()
mexBox.setText("The are no features selected ")
mexBox.exec_()
return
else:
self.g.stopGraph()
self.g.resumeGraph()
self.AddLabelButton.setEnabled(not checked)
self.SelectFeaturesButton.setEnabled(not checked)
for o in self.fixableOperators:
o.inputs["fixAtCurrent"].setValue(not checked)
self.labelListModel.allowRemove(not checked)
self.editor.scheduleSlicesRedraw()
def changeInteractionMode( self, index ):
modes = {0: "navigation", 1: "brushing"}
self.editor.setInteractionMode( modes[index] )
self.interactionComboBox.setCurrentIndex(index)
print "interaction mode switched to", modes[index]
def switchLabel(self, row):
print "switching to label=%r" % (self.labelListModel[row])
#+1 because first is transparent
#FIXME: shouldn't be just row+1 here
self.editor.brushingModel.setDrawnNumber(row+1)
self.editor.brushingModel.setBrushColor(self.labelListModel[row].color)
def switchColor(self, row, color):
print "label=%d changes color to %r" % (row, color)
self.labellayer.colorTable[row]=color.rgba()
self.editor.brushingModel.setBrushColor(color)
self.editor.scheduleSlicesRedraw()
def addLabel(self):
color = QColor(numpy.random.randint(0,255), numpy.random.randint(0,255), numpy.random.randint(0,255))
numLabels = len(self.labelListModel)
if numLabels < len(self._colorTable16):
color = self._colorTable16[numLabels]
self.labellayer.colorTable.append(color.rgba())
self.labelListModel.insertRow(self.labelListModel.rowCount(), Label("Label %d" % (self.labelListModel.rowCount() + 1), color))
nlabels = self.labelListModel.rowCount()
if self.workflow is not None:
print "Label added, changing predictions"
#re-train the forest now that we have more labels
self.workflow.predict.inputs['LabelsCount'].setValue(nlabels)
self.addPredictionLayer(nlabels-1, self.labelListModel._labels[nlabels-1])
#make the new label selected
index = self.labelListModel.index(nlabels-1, 1)
self.labelListModel._selectionModel.select(index, QItemSelectionModel.ClearAndSelect)
#FIXME: this should watch for model changes
#drawing will be enabled when the first label is added
self.changeInteractionMode( 1 )
self.interactionComboBox.setEnabled(True)
def onLabelAboutToBeRemoved(self, parent, start, end):
#the user deleted a label, reshape prediction and remove the layer
#the interface only allows to remove one label at a time?
nout = start-end+1
ncurrent = self.labelListModel.rowCount()
print "removing", nout, "out of ", ncurrent
if self.workflow is not None:
self.workflow.predict.inputs['LabelsCount'].setValue(ncurrent-nout)
for il in range(start, end+1):
labelvalue = self.labelListModel._labels[il]
self.removePredictionLayer(labelvalue)
self.workflow.labels.inputs["deleteLabel"].setValue(il+1)
self.editor.scheduleSlicesRedraw()
def startClassification(self):
nclasses = self.labelListModel.rowCount()
self.workflow.predict.inputs['LabelsCount'].setValue(nclasses)
#add prediction results for all classes as separate channels
for icl in range(nclasses):
self.addPredictionLayer(icl, self.labelListModel._labels[icl])
self.checkInteractive.setEnabled(True)
def addPredictionLayer(self, icl, ref_label):
selector=OpSingleChannelSelector(self.g)
selector.inputs["Input"].connect(self.workflow.prediction_cache.outputs['Output'])
selector.inputs["Index"].setValue(icl)
self.workflow.prediction_cache.inputs["fixAtCurrent"].setValue(not self.checkInteractive.isChecked())
predictsrc = LazyflowSource(selector.outputs["Output"][0])
def srcName(newName):
predictsrc.setObjectName("Prediction for %s" % ref_label.name)
srcName("")
predictLayer = AlphaModulatedLayer(predictsrc, tintColor=ref_label.color, normalize = None )
predictLayer.nameChanged.connect(srcName)
def setLayerColor(c):
print "as the color of label '%s' has changed, setting layer's '%s' tint color to %r" % (ref_label.name, predictLayer.name, c)
predictLayer.tintColor = c
ref_label.colorChanged.connect(setLayerColor)
def setLayerName(n):
newName = "Prediction for %s" % ref_label.name
print "as the name of label '%s' has changed, setting layer's '%s' name to '%s'" % (ref_label.name, predictLayer.name, newName)
predictLayer.name = newName
setLayerName(ref_label.name)
ref_label.nameChanged.connect(setLayerName)
predictLayer.ref_object = ref_label
#make sure that labels (index = 0) stay on top!
self.layerstack.insert(1, predictLayer )
self.fixableOperators.append(self.workflow.prediction_cache)
def removePredictionLayer(self, ref_label):
for il, layer in enumerate(self.layerstack):
if layer.ref_object==ref_label:
print "found the prediction", layer.ref_object, ref_label
self.layerstack.removeRows(il, 1)
break
def openImageStack(self):
self.stackLoader = StackLoader()
self.stackLoader.show()
self.stackLoader.loadButton.clicked.connect(self._stackLoad)
def openFile(self):
fileNames = QFileDialog.getOpenFileNames(self, "Open Image", os.path.abspath(__file__), "Numpy and h5 files (*.npy *.h5)")
if fileNames.count() == 0:
return
self._openFile(fileNames)
def _stackLoad(self):
self.inputProvider = OpArrayPiper(self.g)
op5ifyer = Op5ifyer(self.g)
op5ifyer.inputs["Input"].connect(self.stackLoader.ChainBuilder.outputs["output"])
self.raw = op5ifyer.outputs["Output"][:].allocate().wait()
self.raw = self.raw.view(vigra.VigraArray)
self.min, self.max = numpy.min(self.raw), numpy.max(self.raw)
self.raw.axistags = vigra.AxisTags(
vigra.AxisInfo('t',vigra.AxisType.Time),
vigra.AxisInfo('x',vigra.AxisType.Space),
vigra.AxisInfo('y',vigra.AxisType.Space),
vigra.AxisInfo('z',vigra.AxisType.Space),
vigra.AxisInfo('c',vigra.AxisType.Channels))
self.inputProvider.inputs["Input"].setValue(self.raw)
self.haveData.emit()
self.stackLoader.close()
def _openFile(self, fileNames):
self.inputProvider = None
fName, fExt = os.path.splitext(str(fileNames[0]))
print "Opening Files %r" % fileNames
if fExt=='.npy':
fileName = fileNames[0]
if len(fileNames)>1:
print "WARNING: only the first file will be read, multiple file prediction not supported yet"
fName, fExt = os.path.splitext(str(fileName))
self.raw = numpy.load(str(fileName))
self.min, self.max = numpy.min(self.raw), numpy.max(self.raw)
self.inputProvider = OpArrayPiper(self.g)
self.raw = self.raw.view(vigra.VigraArray)
self.raw.axistags = vigra.AxisTags(
vigra.AxisInfo('t',vigra.AxisType.Time),
vigra.AxisInfo('x',vigra.AxisType.Space),
vigra.AxisInfo('y',vigra.AxisType.Space),
vigra.AxisInfo('z',vigra.AxisType.Space),
vigra.AxisInfo('c',vigra.AxisType.Channels))
self.inputProvider.inputs["Input"].setValue(self.raw)
elif fExt=='.h5':
readerNew=OpH5ReaderBigDataset(self.g)
readerNew.inputs["Filenames"].setValue(fileNames)
readerNew.inputs["hdf5Path"].setValue("volume/data")
readerCache = OpSlicedBlockedArrayCache(self.g)
readerCache.inputs["fixAtCurrent"].setValue(False)
readerCache.inputs["innerBlockShape"].setValue(((1,256,256,1,2),(1,256,1,256,2),(1,1,256,256,2)))
readerCache.inputs["outerBlockShape"].setValue(((1,256,256,4,2),(1,256,4,256,2),(1,4,256,256,2)))
readerCache.inputs["Input"].connect(readerNew.outputs["Output"])
self.inputProvider = OpArrayPiper(self.g)
self.inputProvider.inputs["Input"].connect(readerCache.outputs["Output"])
else:
raise RuntimeError("opening filenames=%r not supported yet" % fileNames)
self.haveData.emit()
def initGraph(self):
shape = self.inputProvider.outputs["Output"].shape
srcs = []
minMax = []
print "* Data has shape=%r" % (shape,)
#create a layer for each channel of the input:
slicer=OpMultiArraySlicer2(self.g)
slicer.inputs["Input"].connect(self.inputProvider.outputs["Output"])
slicer.inputs["AxisFlag"].setValue('c')
nchannels = shape[-1]
for ich in xrange(nchannels):
if self._normalize_data:
data=slicer.outputs['Slices'][ich][:].allocate().wait()
#find the minimum and maximum value for normalization
mm = (numpy.min(data), numpy.max(data))
print " - channel %d: min=%r, max=%r" % (ich, mm[0], mm[1])
minMax.append(mm)
else:
minMax.append(None)
layersrc = LazyflowSource(slicer.outputs['Slices'][ich], priority = 100)
layersrc.setObjectName("raw data channel=%d" % ich)
srcs.append(layersrc)
#FIXME: we shouldn't merge channels automatically, but for now it's prettier
layer1 = None
if nchannels == 1:
layer1 = GrayscaleLayer(srcs[0], normalize=minMax[0])
layer1.set_range(0,minMax[0])
print " - showing raw data as grayscale"
elif nchannels==2:
layer1 = RGBALayer(red = srcs[0], normalizeR=minMax[0],
green = srcs[1], normalizeG=minMax[1])
layer1.set_range(0, minMax[0])
layer1.set_range(1, minMax[1])
print " - showing channel 1 as red, channel 2 as green"
elif nchannels==3:
layer1 = RGBALayer(red = srcs[0], normalizeR=minMax[0],
green = srcs[1], normalizeG=minMax[1],
blue = srcs[2], normalizeB=minMax[2])
layer1.set_range(0, minMax[0])
layer1.set_range(1, minMax[1])
layer1.set_range(2, minMax[2])
print " - showing channel 1 as red, channel 2 as green, channel 3 as blue"
else:
print "only 1,2 or 3 channels supported so far"
return
print
layer1.name = "Input data"
layer1.ref_object = None
self.layerstack.append(layer1)
self.workflow = PixelClassificationLazyflow( self.g, self.featScalesList, self.inputProvider.outputs["Output"])
self.initLabels()
self.startClassification()
self.dataReadyToView.emit()
def initLabels(self):
#Add the layer to draw the labels, but don't add any labels
self.labelsrc = LazyflowSinkSource(self.workflow.labels, self.workflow.labels.outputs["Output"], self.workflow.labels.inputs["Input"])
self.labelsrc.setObjectName("labels")
transparent = QColor(0,0,0,0)
self.labellayer = ColortableLayer(self.labelsrc, colorTable = [transparent.rgba()] )
self.labellayer.name = "Labels"
self.labellayer.ref_object = None
self.layerstack.append(self.labellayer)
def initEditor(self):
shape=self.inputProvider.outputs["Output"].shape
self.editor = VolumeEditor(self.layerstack, labelsink=self.labelsrc)
self.editor.newImageView2DFocus.connect(self.setIconToViewMenu)
#drawing will be enabled when the first label is added
self.editor.setInteractionMode( 'navigation' )
self.volumeEditorWidget.init(self.editor)
model = self.editor.layerStack
self.layerWidget.init(model)
self.UpButton.clicked.connect(model.moveSelectedUp)
model.canMoveSelectedUp.connect(self.UpButton.setEnabled)
self.DownButton.clicked.connect(model.moveSelectedDown)
model.canMoveSelectedDown.connect(self.DownButton.setEnabled)
self.DeleteButton.clicked.connect(model.deleteSelected)
model.canDeleteSelected.connect(self.DeleteButton.setEnabled)
self.workflow.labels.inputs["eraser"].setValue(self.editor.brushingModel.erasingNumber)
#finally, setup the editor to have the correct shape
#doing this last ensures that all connections are setup already
self.editor.dataShape = shape
def _createDefault16ColorColorTable(self):
c = []
c.append(QColor(0, 0, 255))
c.append(QColor(255, 255, 0))
c.append(QColor(255, 0, 0))
c.append(QColor(0, 255, 0))
c.append(QColor(0, 255, 255))
c.append(QColor(255, 0, 255))
c.append(QColor(255, 105, 180)) #hot pink
c.append(QColor(102, 205, 170)) #dark aquamarine
c.append(QColor(165, 42, 42)) #brown
c.append(QColor(0, 0, 128)) #navy
c.append(QColor(255, 165, 0)) #orange
c.append(QColor(173, 255, 47)) #green-yellow
c.append(QColor(128,0, 128)) #purple
c.append(QColor(192, 192, 192)) #silver
c.append(QColor(240, 230, 140)) #khaki
c.append(QColor(69, 69, 69)) # dark grey
return c
def onFeatureButtonClicked(self):
self.featureDlg.show()
def _onFeaturesChosen(self):
selectedFeatures = self.featureDlg.featureTableWidget.createSelectedFeaturesBoolMatrix()
print "new feature set:", selectedFeatures
self.workflow.features.inputs['Matrix'].setValue(numpy.asarray(selectedFeatures))
def _initFeatureDlg(self):
dlg = self.featureDlg = FeatureDlg()
dlg.setWindowTitle("Features")
dlg.createFeatureTable({"Features": [FeatureEntry("Gaussian smoothing"), \
FeatureEntry("Laplacian of Gaussian"), \
FeatureEntry("Structure Tensor Eigenvalues"), \
FeatureEntry("Hessian of Gaussian EV"), \
FeatureEntry("Gaussian Gradient Magnitude"), \
FeatureEntry("Difference Of Gaussian")]}, \
self.featScalesList)
dlg.setImageToPreView(None)
m = [[1,0,0,0,0,0,0],[1,0,0,0,0,0,0],[0,0,0,0,0,0,0],[1,0,0,0,0,0,0],[1,0,0,0,0,0,0],[1,0,0,0,0,0,0]]
dlg.featureTableWidget.setSelectedFeatureBoolMatrix(m)
dlg.accepted.connect(self._onFeaturesChosen)
def openImageStack(self):
self.stackLoader = StackLoader()
self.stackLoader.show()
self.stackLoader.loadButton.clicked.connect(self._stackLoad)
