def __init__(self, slot, operator, inslot=None, name=None,
subname=None, default=None, depends=None,
selfdepends=True):
super(SerialBoxSlot, self).__init__(
slot, inslot, name, subname, default, depends, selfdepends
)
self.operator = operator
self.progressSignal = OrderedSignal() # Signature: __call__(percentComplete)
def __init__( self, name, syncWithImageIndex=True, interactive=True ):
"""
Constructor.
Subclasses must call this base implementation in their own ``__init__`` methods.
If they fail to do so, the shell raises an exception.
:param name: The applet's name, which will appear as the applet drawer title.
:param syncWithImageIndex: If True, the shell/workflow will add an image lane to this applet for each image in the interactive workflow.
:param interactive: If False, the applet controls won't be shown in the applet bar GUI.
"""
self.name = name
self.syncWithImageIndex = syncWithImageIndex
self.__interactive = interactive
self.busy = False
#: Progress signal.
#: When the applet is doing something time-consuming, this signal tells the shell to show a progress bar.
#: Signature: ``__call__(percentComplete, canceled=False)``
#:
#: .. note:: To update the progress bar correctly, the shell expects that progress updates always
#: begin with at least one zero update and end with at least one 100 update.
#: That is:
#: ``self.progressSignal(0)`` ... more updates ... ``self.progressSignal(100)``
self.progressSignal = OrderedSignal()
#: Shell request signal is used to trigger certain shell actions.
#: Signature: ``__call__(request)``
#: where ``request`` is an integer corresponding to the action the shell should take.
#: The allowable actions are enumerated in the :py:class:`ShellRequest` class.
#: Example invocation: ``self.shellRequest(ShellRequest.RequestSave)``
self.shellRequestSignal = OrderedSignal()
#: This signal informs the workflow that something has changed that might
#: affect the usability of various applets in the workflow.
#: Signature: ``emit()``
self.appletStateUpdateRequested = OrderedSignal()
#: This signal tells the shell to send the dict 'data' to the (TCP) server
#: 'name' (if connected)
#: Signature: ``__call__(servername, data)``
self.sendMessageToServer = OrderedSignal()
self._base_initialized = True
def __init__(self, slot, operator, inslot=None, name=None,
subname=None, default=None, depends=None,
selfdepends=True):
super(SerialPredictionSlot, self).__init__(
slot, inslot, name, subname, default, depends, selfdepends
)
self.operator = operator
self.progressSignal = OrderedSignal()
self._predictionStorageEnabled = False
self._predictionStorageRequest = None
self._predictionsPresent = False
def __init__(self, topGroupName, slots=None, operator=None):
"""Constructor. Subclasses must call this method in their own
__init__ functions. If they fail to do so, the shell raises an
exception.
Parameters:
:param topGroupName: name of this applet's data group in the file.
Defaults to the name of the operator.
:param slots: a list of SerialSlots
"""
self.progressSignal = OrderedSignal(
) # Signature: __call__(percentComplete)
self.base_initialized = True
self.topGroupName = topGroupName
self.serialSlots = maybe(slots, [])
self.operator = operator
self._ignoreDirty = False
class EdgeTrainingMixin:
DEFAULT_PEN = QPen(SegmentationEdgesLayer.DEFAULT_PEN)
DEFAULT_PEN.setColor(Qt.yellow)
# signal used to synchronize live update button enable status across lanes
labels_updated = OrderedSignal()
###########################################
### AppletGuiInterface Concrete Methods ###
###########################################
def appletDrawer(self):
return self._drawer
def stopAndCleanUp(self):
# Unsubscribe to all signals
for fn in self.__cleanup_fns:
fn()
# Base class
super().stopAndCleanUp()
###########################################
###########################################
def __init_subclass__(cls, **kwargs):
"""Make sure Mixin can only be used with LayerViewerGui"""
assert issubclass(
cls,
LayerViewerGui), "Mixin should only be used with LayerViewerGui"
super().__init_subclass__(**kwargs)
def __init__(self, parentApplet, topLevelOperatorView, **kwargs):
self._currently_updating = False
self.__cleanup_fns = []
self.parentApplet = parentApplet
self.topLevelOperatorView = topLevelOperatorView
super().__init__(parentApplet, topLevelOperatorView, **kwargs)
self._init_edge_label_colortable()
self._init_probability_colortable()
def _after_init(self):
super()._after_init()
self.update_probability_edges()
# Initialize everything with the operator's initial values
self.configure_gui_from_operator()
# need to call this callback manually, as signals a re blocked in
# configure_gui_from_operator()
self._handle_live_update_clicked(self.live_update_button.isChecked())
def createDrawerControls(self):
op = self.topLevelOperatorView
def configure_update_handlers(qt_signal, op_slot):
qt_signal.connect(self.configure_operator_from_gui)
cleanup_fn = op_slot.notifyDirty(self.configure_gui_from_operator,
defer=True)
self.__cleanup_fns.append(cleanup_fn)
# Controls
feature_selection_button = QPushButton(
text="Select Features",
icon=QIcon(ilastikIcons.AddSel),
toolTip=
"Select edge/superpixel features to use for classification.",
clicked=self._open_feature_selection_dlg,
)
self.train_from_gt_button = QPushButton(
text="Auto-label",
icon=QIcon(ilastikIcons.Segment),
toolTip=
"Automatically label all edges according to your pre-loaded groundtruth volume.",
clicked=self._handle_label_from_gt_clicked,
)
self.clear_labels_button = QPushButton(
text="Clear Labels",
icon=QIcon(ilastikIcons.Clear),
toolTip="Remove all edge labels. (Start over on this image.)",
clicked=self._handle_clear_labels_clicked,
)
self.live_update_button = QPushButton(
text="Live Predict",
checkable=True,
icon=QIcon(ilastikIcons.Play),
toolTip="Update the edge classifier predictions",
clicked=self._handle_live_update_clicked,
enabled=False,
)
configure_update_handlers(self.live_update_button.toggled,
op.FreezeClassifier)
configure_update_handlers(self.train_from_gt_button.toggled,
op.TrainRandomForest)
cleanup_fn = op.EdgeLabelsDict.notifyDirty(
self.any_edge_annotations_available)
self.__cleanup_fns.append(cleanup_fn)
# call once when instantiating with a saved project to make the live update button available
# if there are annotations loaded from file.
self.labels_updated.subscribe(self.enable_live_update_button)
self.__cleanup_fns.append(
partial(self.labels_updated.unsubscribe,
self.enable_live_update_button))
self.any_edge_annotations_available()
# Layout
label_layout = QHBoxLayout()
label_layout.addWidget(self.clear_labels_button)
label_layout.addWidget(self.train_from_gt_button)
label_layout.setSpacing(1)
layout = QVBoxLayout()
layout.addWidget(feature_selection_button)
layout.setSpacing(1)
layout.addLayout(label_layout)
layout.addWidget(self.live_update_button)
layout.addSpacerItem(
QSpacerItem(0, 10, QSizePolicy.Minimum, QSizePolicy.Expanding))
# Finally, the whole drawer widget
drawer = QWidget(parent=self)
drawer.setLayout(layout)
# Widget Shortcuts
mgr = ShortcutManager()
ActionInfo = ShortcutManager.ActionInfo
shortcut_group = "Edge Training"
mgr.register(
"l",
ActionInfo(
shortcut_group,
"Live Predict",
"Toggle live edge classifier update mode",
self.live_update_button.toggle,
self.live_update_button,
self.live_update_button,
),
)
return drawer
def any_edge_annotations_available(self, *args, **kwargs):
any_have_edges = False
op = self.topLevelOperatorView
top_level_edge_labels_dict = op.EdgeLabelsDict.top_level_slot
assert top_level_edge_labels_dict.level == 1
for subslot in op.EdgeLabelsDict.top_level_slot:
any_have_edges = subslot.ready() and bool(subslot.value)
if any_have_edges:
break
self.labels_updated(any_have_edges)
@threadRouted
def enable_live_update_button(self, enable):
self.live_update_button.setEnabled(enable)
def initAppletDrawerUi(self):
"""
Overridden from base class (LayerViewerGui)
"""
op = self.topLevelOperatorView
cleanup_fn = op.GroundtruthSegmentation.notifyReady(
self.configure_gui_from_operator, defer=True)
self.__cleanup_fns.append(cleanup_fn)
def _open_feature_selection_dlg(self):
rag = self.topLevelOperatorView.Rag.value
feature_names = rag.supported_features()
channel_names = self.topLevelOperatorView.VoxelData.meta.channel_names
default_selections = self.topLevelOperatorView.FeatureNames.value
def decodeToStringIfBytes(s):
if isinstance(s, bytes):
return s.decode()
else:
return s
channel_names = [decodeToStringIfBytes(s) for s in channel_names]
feature_names = [decodeToStringIfBytes(s) for s in feature_names]
# default_selections
# *dict, str: list - of - str *
default_selections_strings = {}
for key, value in default_selections.items():
default_selections_strings[decodeToStringIfBytes(key)] = [
decodeToStringIfBytes(s) for s in value
]
dlg = FeatureSelectionDialog(channel_names,
feature_names,
default_selections_strings,
parent=self)
dlg_result = dlg.exec_()
if dlg_result != dlg.Accepted:
return
selections = dlg.selections()
self.topLevelOperatorView.FeatureNames.setValue(selections)
# Configure the handler for updated edge label maps
def _init_edge_label_colortable(self):
self.edge_label_colortable = [
QColor(0, 255, 0, 255), # green
QColor(255, 0, 0, 255),
] # red
self.edge_label_pen_table = [
self.DEFAULT_PEN,
]
for color in self.edge_label_colortable:
pen = QPen(SegmentationEdgesLayer.DEFAULT_PEN)
pen.setColor(color)
pen.setWidth(5)
self.edge_label_pen_table.append(pen)
# When the edge labels are dirty, update the edge label layer pens
op = self.topLevelOperatorView
cleanup_fn = op.EdgeLabelsDict.notifyDirty(self.update_labeled_edges,
defer=True)
self.__cleanup_fns.append(cleanup_fn)
@threadRouted
def update_labeled_edges(self, *args):
op = self.topLevelOperatorView
edge_label_layer = self.getLayerByName("Edge Labels")
if not edge_label_layer:
return
edge_label_layer.overwrite_edge_labels(op.EdgeLabelsDict.value)
def _handle_edge_label_clicked(self, updated_edge_labels):
"""
The user clicked an edge label.
Update the operator with the new values.
"""
op = self.topLevelOperatorView
edge_labels = op.EdgeLabelsDict.value
new_labels = dict(edge_labels)
new_labels.update(updated_edge_labels)
for sp_id_pair, new_label in dict(new_labels).items():
if new_label == 0:
del new_labels[sp_id_pair]
op.EdgeLabelsDict.setValue(new_labels)
def _handle_label_from_gt_clicked(self):
def train_from_gt():
try:
op = self.topLevelOperatorView
op.setEdgeLabelsFromGroundtruth(op.current_view_index())
finally:
self.parentApplet.busy = False
self.parentApplet.progressSignal(100)
self.parentApplet.appletStateUpdateRequested()
self.parentApplet.busy = True
self.parentApplet.progressSignal(-1)
self.parentApplet.appletStateUpdateRequested()
Request(train_from_gt).submit()
def _handle_clear_labels_clicked(self):
response = QMessageBox.warning(
self,
"Clear Labels?",
"This will clear all edge labels in the current image.\nAre you sure?",
buttons=QMessageBox.Ok | QMessageBox.Cancel,
)
if response == QMessageBox.Ok:
op = self.topLevelOperatorView
op.EdgeLabelsDict.setValue({})
op.FreezeClassifier.setValue(True)
def _handle_live_update_clicked(self, checked):
if checked:
probs_layer = self.getLayerByName("Edge Probabilities")
if probs_layer:
probs_layer.visible = True
# Configure the handler for updated probability maps
# FIXME: Should we make a new Layer subclass that handles this colortable mapping for us? Yes.
def _init_probability_colortable(self):
self.probability_colortable = []
for v in np.linspace(0.0, 1.0, num=101):
self.probability_colortable.append(
QColor(255 * (v), 255 * (1.0 - v), 0))
self.probability_pen_table = []
for color in self.probability_colortable:
pen = QPen(SegmentationEdgesLayer.DEFAULT_PEN)
pen.setColor(color)
self.probability_pen_table.append(pen)
# When the edge probabilities are dirty, update the probability edge layer pens
op = self.topLevelOperatorView
cleanup_fn = op.EdgeProbabilitiesDict.notifyDirty(
self.update_probability_edges, defer=True)
self.__cleanup_fns.append(cleanup_fn)
def update_probability_edges(self, *args):
def _impl():
op = self.topLevelOperatorView
if not self.getLayerByName("Edge Probabilities"):
return
edge_probs = op.EdgeProbabilitiesDict.value
new_pens = {}
for id_pair, probability in list(edge_probs.items()):
new_pens[id_pair] = self.probability_pen_table[int(
probability * 100)]
self.apply_new_probability_edges(new_pens)
# submit the worklaod in a request and return immediately
req = Request(_impl).submit()
# Now that we've trained the classifier, the workflow may wish to enable downstream applets.
self.parentApplet.appletStateUpdateRequested()
@threadRouted
def apply_new_probability_edges(self, new_pens):
# This function is threadRouted because you can't
# touch the layer colortable outside the main thread.
superpixel_edge_layer = self.getLayerByName("Edge Probabilities")
if superpixel_edge_layer:
superpixel_edge_layer.pen_table.overwrite(new_pens)
@contextmanager
def set_updating(self):
assert not self._currently_updating
self._currently_updating = True
try:
yield
finally:
self._currently_updating = False
def configure_gui_from_operator(self, *args):
if self._currently_updating:
return False
with self.set_updating():
op = self.topLevelOperatorView
with silent_qobject(self.train_from_gt_button) as w:
w.setEnabled(op.GroundtruthSegmentation.ready())
with silent_qobject(self.live_update_button) as w:
w.setChecked(not op.FreezeClassifier.value)
if op.FreezeClassifier.value:
self.live_update_button.setIcon(QIcon(ilastikIcons.Play))
else:
self.live_update_button.setIcon(QIcon(ilastikIcons.Pause))
def configure_operator_from_gui(self):
if self._currently_updating:
return False
with self.set_updating():
op = self.topLevelOperatorView
op.FreezeClassifier.setValue(
not self.live_update_button.isChecked())
def create_prefetch_menu(self, layer_name):
def prefetch_layer(axis="z"):
layer_index = self.layerstack.findMatchingIndex(
lambda l: l.name == layer_name)
num_slices = self.editor.dataShape["txyzc".index(axis)]
view2d = self.editor.imageViews["xyz".index(axis)]
view2d.scene().triggerPrefetch([layer_index],
spatial_axis_range=(0, num_slices))
prefetch_menu = QMenu("Prefetch")
prefetch_menu.addAction(
QAction("All Z-slices",
prefetch_menu,
triggered=partial(prefetch_layer, "z")))
prefetch_menu.addAction(
QAction("All Y-slices",
prefetch_menu,
triggered=partial(prefetch_layer, "y")))
prefetch_menu.addAction(
QAction("All X-slices",
prefetch_menu,
triggered=partial(prefetch_layer, "x")))
return prefetch_menu
def setupLayers(self):
layers = []
op = self.topLevelOperatorView
ActionInfo = ShortcutManager.ActionInfo
superpixels_ready = op.Superpixels.ready()
with_training = op.TrainRandomForest.value
# Superpixels -- Edge Probabilities
if superpixels_ready and op.EdgeProbabilitiesDict.ready(
) and with_training:
layer = SegmentationEdgesLayer(createDataSource(op.Superpixels),
isHoverable=True)
layer.name = "Edge Probabilities" # Name is hard-coded in multiple places: grep before changing.
layer.visible = False
layer.opacity = 1.0
self.update_probability_edges() # Initialize
layer.contexts.append(
self.create_prefetch_menu("Edge Probabilities"))
layer.shortcutRegistration = (
"p",
ActionInfo(
"Edge Training Layers",
"EdgePredictionsVisibility",
"Show/Hide Edge Predictions",
layer.toggleVisible,
self.viewerControlWidget(),
layer,
),
)
layers.append(layer)
del layer
# Superpixels -- Edge Labels
if superpixels_ready and op.EdgeLabelsDict.ready() and with_training:
edge_labels = op.EdgeLabelsDict.value
layer = LabelableSegmentationEdgesLayer(
createDataSource(op.Superpixels), self.edge_label_pen_table,
edge_labels)
layer.name = "Edge Labels"
layer.visible = True
layer.opacity = 1.0
self.update_labeled_edges() # Initialize
layer.labelsChanged.connect(self._handle_edge_label_clicked)
layer.contexts.append(self.create_prefetch_menu("Edge Labels"))
layer.shortcutRegistration = (
"0",
ActionInfo(
"Edge Training Layers",
"LabelVisibility",
"Show/Hide Edge Labels",
layer.toggleVisible,
self.viewerControlWidget(),
layer,
),
)
layers.append(layer)
del layer
# Superpixels -- Edges
if superpixels_ready:
layer = SegmentationEdgesLayer(createDataSource(op.Superpixels),
default_pen=self.DEFAULT_PEN,
isHoverable=with_training)
layer.name = "Superpixel Edges"
layer.visible = True
layer.opacity = 1.0
layers.append(layer)
del layer
# Naive Segmentation
if op.NaiveSegmentation.ready():
layer = self.createStandardLayerFromSlot(op.NaiveSegmentation)
layer.name = "Naive Segmentation"
layer.visible = False
layer.opacity = 0.5
layer.shortcutRegistration = (
"n",
ActionInfo(
"Edge Training Layers",
"NaiveSegmentationVisibility",
"Show/Hide Naive Segmentation (shows output if classifier output is respected verbatim)",
layer.toggleVisible,
self.viewerControlWidget(),
layer,
),
)
layers.append(layer)
del layer
# Groundtruth
if op.GroundtruthSegmentation.ready():
layer = self.createStandardLayerFromSlot(
op.GroundtruthSegmentation)
layer.name = "Groundtruth"
layer.visible = False
layer.opacity = 0.5
layer.shortcutRegistration = (
"g",
ActionInfo(
"Edge Training Layers",
"GroundtruthVisibility",
"Show/Hide Groundtruth",
layer.toggleVisible,
self.viewerControlWidget(),
layer,
),
)
layers.append(layer)
del layer
# Voxel data
if op.VoxelData.ready():
layer = self._create_grayscale_layer_from_slot(
op.VoxelData,
op.VoxelData.meta.getTaggedShape()["c"])
layer.name = "Voxel Data"
layer.visible = False
layer.opacity = 1.0
layers.append(layer)
del layer
# Raw Data (grayscale)
if op.RawData.ready():
layer = self.createStandardLayerFromSlot(op.RawData)
layer.name = "Raw Data"
layer.visible = True
layer.opacity = 1.0
layers.append(layer)
layer.shortcutRegistration = (
"i",
ActionInfo(
"Edge Training Layers",
"Hide all but Raw",
"Hide all but Raw",
partial(self.toggle_show_raw, "Raw Data"),
self.viewerControlWidget(),
layer,
),
)
del layer
return layers