class OpGraphCutSegmentation(Operator):
RawInput = InputSlot() # FIXME is this neccessary?
InputImage = InputSlot()
LabelImage = InputSlot()
Beta = InputSlot(value=.2)
Channel = InputSlot(value=0)
CachedOutput = OutputSlot()
def __init__(self, *args, **kwargs):
super(OpGraphCutSegmentation, self).__init__(*args, **kwargs)
op = OpObjectsSegment(parent=self)
op.Prediction.connect(self.InputImage)
op.LabelImage.connect(self.LabelImage)
op.Beta.connect(self.Beta)
op.Channel.connect(self.Channel)
self.CachedOutput.connect(op.CachedOutput)
self._op = op
self._filled = False
def setupOutputs(self):
pass
def propagateDirty(self, slot, subindex, roi):
pass # Nothing to do
def execute(self, slot, subindex, roi, result):
assert False, "Shuld not get here"
class DirtyAssert(Operator):
Input = InputSlot()
def propagateDirty(self, slot, subindex, roi):
assert np.all(roi.start == 0)
assert np.all(roi.stop == self.Input.meta.shape)
raise PropagateDirtyCalled()
class OpFillMaskArray(Operator):
name = "OpFillMaskArray"
category = "Pointwise"
InputArray = InputSlot(allow_mask=True)
InputFillValue = InputSlot(optional=True)
Output = OutputSlot()
def __init__(self, *args, **kwargs):
super(OpFillMaskArray, self).__init__(*args, **kwargs)
def setupOutputs(self):
# Copy the input metadata to both outputs
self.Output.meta.assignFrom(self.InputArray.meta)
self.Output.meta.has_mask = False
def execute(self, slot, subindex, roi, result):
key = roi.toSlice()
# Get data
data = self.InputArray[key].wait()
# Copy results
if slot.name == 'Output':
if not isinstance(data, numpy.ma.masked_array):
result[...] = data
elif self.InputFillValue.ready():
result[...] = data.filled(self.InputFillValue.value)
else:
result[...] = data.filled()
def propagateDirty(self, slot, subindex, roi):
if (slot.name == "InputArray"):
slicing = roi.toSlice()
self.Output.setDirty(slicing)
elif (slot.name == "InputFillValue"):
self.Output.setDirty(slice(None))
else:
assert False, "Unknown dirty input slot"
class DirtyAssert(Operator):
Input = InputSlot()
def willBeDirty(self, t, c):
self._t = t
self._c = c
def propagateDirty(self, slot, subindex, roi):
t_ind = self.Input.meta.axistags.index("t")
c_ind = self.Input.meta.axistags.index("c")
assert roi.start[t_ind] == self._t
assert roi.start[c_ind] == self._c
assert roi.stop[t_ind] == self._t + 1
assert roi.stop[c_ind] == self._c + 1
raise PropagateDirtyCalled()
class OpMaskArray(Operator):
name = "OpMaskArray"
category = "Pointwise"
InputArray = InputSlot(allow_mask=True)
InputMask = InputSlot()
Output = OutputSlot(allow_mask=True)
def __init__(self, *args, **kwargs):
super(OpMaskArray, self).__init__(*args, **kwargs)
def setupOutputs(self):
# Copy the input metadata to both outputs
self.Output.meta.assignFrom(self.InputArray.meta)
self.Output.meta.has_mask = True
def execute(self, slot, subindex, roi, result):
key = roi.toSlice()
if slot.name == 'Output':
# Write data into result (including a mask if provided)
self.InputArray[key].writeInto(result).wait()
# Get the added mask
mask = self.InputMask[key].wait()
# Apply the combination of the masks to result.
result.mask[...] |= mask
def propagateDirty(self, slot, subindex, roi):
if (slot.name == "InputArray") or (slot.name == "InputMask"):
slicing = roi.toSlice()
self.Output.setDirty(slicing)
else:
assert False, "Unknown dirty input slot"
class CountExecutes(Operator):
Input = InputSlot()
Output = OutputSlot()
numExecutes = 0
def setupOutputs(self):
self.Output.meta.assignFrom(self.Input.meta)
def propagateDirty(self, slot, subindex, roi):
self.Output.setDirty(roi)
def execute(self, slot, sunbindex, roi, result):
self.numExecutes += 1
req = self.Input.get(roi)
req.writeInto(result)
req.block()
class OpSplitMaskArray(Operator):
name = "OpSplitMaskArray"
category = "Pointwise"
Input = InputSlot(allow_mask=True)
OutputArray = OutputSlot()
OutputMask = OutputSlot()
OutputFillValue = OutputSlot()
def __init__(self, *args, **kwargs):
super( OpSplitMaskArray, self ).__init__( *args, **kwargs )
def setupOutputs(self):
# Copy the input metadata to both outputs
self.OutputArray.meta.assignFrom( self.Input.meta )
self.OutputArray.meta.has_mask = False
self.OutputMask.meta.assignFrom( self.Input.meta )
self.OutputMask.meta.has_mask = False
self.OutputMask.meta.dtype = numpy.bool8
self.OutputFillValue.meta.assignFrom( self.Input.meta )
self.OutputFillValue.meta.has_mask = False
self.OutputFillValue.meta.shape = tuple()
def execute(self, slot, subindex, roi, result):
key = roi.toSlice()
input_subview = self.Input[key].wait()
if slot.name == 'OutputArray':
result[...] = input_subview.data
elif slot.name == 'OutputMask':
result[...] = input_subview.mask
elif slot.name == 'OutputFillValue':
result[...] = input_subview.fill_value
def propagateDirty(self, slot, subindex, roi):
if (slot.name == "Input"):
slicing = roi.toSlice()
self.OutputArray.setDirty(slicing)
self.OutputMask.setDirty(slicing)
self.OutputFillValue.setDirty(Ellipsis)
else:
assert False, "Unknown dirty input slot"
class OpA(Operator):
input = InputSlot()
output = OutputSlot()
def __init__(self, parent=None):
Operator.__init__(self, parent)
self.countSetupOutputs = 0
def setupOutputs(self):
self.countSetupOutputs += 1
self.output.meta.shape = self.input.meta.shape
self.output.meta.dtype = self.input.meta.dtype
def execute(self, slot, subindex, roi, result):
pass
def propagateDirty(self, slot, subindex, roi):
pass
class OpLabelVolume(Operator):
name = "OpLabelVolume"
## provide the volume to label here
# (arbitrary shape, dtype could be restricted, see the implementations
# property supportedDtypes below)
Input = InputSlot()
## provide labels that are treated as background
# the shape of the background labels must match the shape of the volume in
# channel and in time axis, and must have no spatial axes.
# E.g.: volume.taggedShape = {'x': 10, 'y': 12, 'z': 5, 'c': 3, 't': 100}
# ==>
# background.taggedShape = {'c': 3, 't': 100}
# TODO relax requirements (single value is already working)
Background = InputSlot(optional=True)
# Bypass cache (for headless mode)
BypassModeEnabled = InputSlot(value=False)
## decide which CCL method to use
#
# currently available:
# * 'vigra': use the fast algorithm from ukoethe/vigra
# * 'blocked': use the memory saving algorithm from thorbenk/blockedarray
#
# A change here deletes all previously cached results.
Method = InputSlot(value="vigra")
## Labeled volume
# Axistags and shape are the same as on the Input, dtype is an integer
# datatype.
# This slot operates on a what-you-request-is-what-you-get basis, if you
# request a subregion only that subregion will be considered for labeling
# and no internal caches are used. If you want consistent labels for
# subsequent requests, use CachedOutput instead.
# This slot will be set dirty by time and channel if the background or the
# input changes for the respective time-channel-slice.
Output = OutputSlot()
## Cached label image
# Axistags and shape are the same as on the Input, dtype is an integer
# datatype.
# This slot extends the ROI to the full xyz volume (c and t are unaffected)
# and computes the labeling for the whole volume. As long as the input does
# not get dirty, subsequent requests to this slot guarantee consistent
# labelings. The internal cache in use is an OpCompressedCache.
# This slot will be set dirty by time and channel if the background or the
# input changes for the respective time-channel-slice.
CachedOutput = OutputSlot()
# cache access, see OpCompressedCache
CleanBlocks = OutputSlot()
def __init__(self, *args, **kwargs):
super(OpLabelVolume, self).__init__(*args, **kwargs)
# we just want to have 5d data internally
op5 = OpReorderAxes(parent=self)
op5.Input.connect(self.Input)
op5.AxisOrder.setValue("txyzc")
self._op5 = op5
self._opLabel = None
self._op5_2 = OpReorderAxes(parent=self)
self._op5_2_cached = OpReorderAxes(parent=self)
self.Output.connect(self._op5_2.Output)
self.CachedOutput.connect(self._op5_2_cached.Output)
# available OpLabelingABCs:
# TODO: OpLazyConnectedComponents and _OpLabelBlocked does not conform to OpLabelingABC
self._labelOps = {
"vigra": _OpLabelVigra,
"blocked": _OpLabelBlocked,
"lazy": OpLazyConnectedComponents
}
def setupOutputs(self):
method = self.Method.value
if not isinstance(method, (str, unicode)):
method = method[0]
if self._opLabel is not None and type(
self._opLabel) != self._labelOps[method]:
# fully remove old labeling operator
self._op5_2.Input.disconnect()
self._op5_2_cached.Input.disconnect()
self._opLabel.Input.disconnect()
self._opLabel = None
if self._opLabel is None:
self._opLabel = self._labelOps[method](parent=self)
self._opLabel.Input.connect(self._op5.Output)
if method is "vigra":
self._opLabel.BypassModeEnabled.connect(self.BypassModeEnabled)
# connect reordering operators
self._op5_2.Input.connect(self._opLabel.Output)
self._op5_2_cached.Input.connect(self._opLabel.CachedOutput)
# set the final reordering operator's AxisOrder to that of the input
origOrder = self.Input.meta.getAxisKeys()
self._op5_2.AxisOrder.setValue(origOrder)
self._op5_2_cached.AxisOrder.setValue(origOrder)
# connect cache access slots
self.CleanBlocks.connect(self._opLabel.CleanBlocks)
# set background values
self._setBG()
def propagateDirty(self, slot, subindex, roi):
if slot == self.BypassModeEnabled:
pass
elif slot == self.Method:
# We are changing the labeling method. In principle, the labelings
# are equivalent, but not necessarily the same!
self.Output.setDirty(slice(None))
elif slot == self.Input:
# handled by internal operator
pass
elif slot == self.Background:
# propagate the background values, output will be set dirty in
# internal operator
self._setBG()
def setInSlot(self, slot, subindex, roi, value):
# "Invalid slot for setInSlot(): {}".format( slot.name )
# Nothing to do here.
# Our Input slots are directly fed into the cache,
# so all calls to __setitem__ are forwarded automatically
pass
## set the background values of inner operator
def _setBG(self):
if self.Background.ready():
val = self.Background.value
else:
val = 0
bg = np.asarray(val)
t = self._op5.Output.meta.shape[0]
c = self._op5.Output.meta.shape[4]
if bg.size == 1:
bg = np.zeros((c, t))
bg[:] = val
bg = vigra.taggedView(bg, axistags="ct")
else:
bg = vigra.taggedView(val, axistags=self.Background.meta.axistags)
bg = bg.withAxes(*"ct")
bg = bg.withAxes(*"txyzc")
self._opLabel.Background.setValue(bg)
class OpLabelingABC(with_metaclass(ABCMeta, Operator)):
Input = InputSlot()
## background with axes 'txyzc', spatial axes must be singletons
Background = InputSlot()
# Bypass cache (for headless mode)
BypassModeEnabled = InputSlot(value=False)
Output = OutputSlot()
CachedOutput = OutputSlot()
# cache access, see OpCompressedCache
CleanBlocks = OutputSlot()
# the numeric type that is used for labeling
labelType = np.uint32
## list of supported dtypes
@abstractproperty
def supportedDtypes(self):
pass
def __init__(self, *args, **kwargs):
super(OpLabelingABC, self).__init__(*args, **kwargs)
self._cache = OpBlockedArrayCache(parent=self)
self._cache.name = "OpLabelVolume.OutputCache"
self._cache.BypassModeEnabled.connect(self.BypassModeEnabled)
self._cache.Input.connect(self.Output)
self.CachedOutput.connect(self._cache.Output)
self.CleanBlocks.connect(self._cache.CleanBlocks)
def setupOutputs(self):
# check if the input dtype is valid
if self.Input.ready():
dtype = self.Input.meta.dtype
if dtype not in self.supportedDtypes:
msg = "{}: dtype '{}' not supported " "with method 'vigra'. Supported types: {}"
msg = msg.format(self.name, dtype, self.supportedDtypes)
raise ValueError(msg)
# set cache chunk shape to the whole spatial volume
shape = np.asarray(self.Input.meta.shape, dtype=np.int)
shape[0] = 1
shape[4] = 1
self._cache.BlockShape.setValue(tuple(shape))
# setup meta for Output
self.Output.meta.assignFrom(self.Input.meta)
self.Output.meta.dtype = self.labelType
def propagateDirty(self, slot, subindex, roi):
if slot == self.BypassModeEnabled:
pass
else:
# a change in either input or background makes the whole
# time-channel-slice dirty (CCL is a global operation)
outroi = roi.copy()
outroi.start[1:4] = (0, 0, 0)
outroi.stop[1:4] = self.Input.meta.shape[1:4]
self.Output.setDirty(outroi)
self.CachedOutput.setDirty(outroi)
def setInSlot(self, slot, subindex, roi, value):
# "Invalid slot for setInSlot(): {}".format( slot.name )
# Nothing to do here.
# Our Input slots are directly fed into the cache,
# so all calls to __setitem__ are forwarded automatically
pass
def execute(self, slot, subindex, roi, result):
if slot == self.Output:
# just label the ROI and write it to result
self._label(roi, result)
else:
raise ValueError("Request to unknown slot {}".format(slot))
def _label(self, roi, result):
result = vigra.taggedView(result, axistags=self.Output.meta.axistags)
# get the background values
bg = self.Background[...].wait()
bg = vigra.taggedView(bg, axistags=self.Background.meta.axistags)
bg = bg.withAxes(*"ct")
assert np.all(
self.Background.meta.shape[0] == self.Input.meta.shape[0]
), "Shape of background values incompatible to shape of Input"
assert np.all(
self.Background.meta.shape[4] == self.Input.meta.shape[4]
), "Shape of background values incompatible to shape of Input"
# do labeling in parallel over channels and time slices
pool = RequestPool()
start = np.asarray(roi.start, dtype=np.int)
stop = np.asarray(roi.stop, dtype=np.int)
for ti, t in enumerate(range(roi.start[0], roi.stop[0])):
start[0], stop[0] = t, t + 1
for ci, c in enumerate(range(roi.start[4], roi.stop[4])):
start[4], stop[4] = c, c + 1
newRoi = SubRegion(self.Output,
start=tuple(start),
stop=tuple(stop))
resView = result[ti, ..., ci].withAxes(*"xyz")
req = Request(partial(self._label3d, newRoi, bg[c, t],
resView))
pool.add(req)
logger.debug(
"{}: Computing connected components for ROI {} ...".format(
self.name, roi))
pool.wait()
pool.clean()
logger.debug("{}: Connected components computed.".format(self.name))
## compute the requested roi and put the results into result
#
# @param result the array to write into, 3d xyz
@abstractmethod
def _label3d(self, roi, bg, result):
pass
class OpObjectsSegment(OpGraphCut):
name = "OpObjectsSegment"
# thresholded predictions, or otherwise obtained ROI indicators
# (a value of 0 is assumed to be background and ignored)
LabelImage = InputSlot()
# margin around each object (always xyz!)
Margin = InputSlot(value=np.asarray((20, 20, 20)))
# bounding boxes of the labeled objects
# this slot returns an array of dicts with shape (t, c)
BoundingBoxes = OutputSlot(stype=Opaque)
### slots from OpGraphCut ###
## prediction maps
#Prediction = InputSlot()
## graph cut parameter
#Beta = InputSlot(value=.2)
## labeled segmentation image
#Output = OutputSlot()
#CachedOutput = OutputSlot()
def __init__(self, *args, **kwargs):
super(OpObjectsSegment, self).__init__(*args, **kwargs)
def setupOutputs(self):
super(OpObjectsSegment, self).setupOutputs()
# sanity checks
shape = self.LabelImage.meta.shape
assert len(shape) == 5,\
"Prediction maps must be a full 5d volume (txyzc)"
tags = self.LabelImage.meta.getAxisKeys()
tags = "".join(tags)
assert tags == 'txyzc',\
"Label image has wrong axes order"\
"(expected: txyzc, got: {})".format(tags)
# bounding boxes are just one element arrays of type object, but we
# want to request boxes from a specific region, therefore BoundingBoxes
# needs a shape
shape = self.Prediction.meta.shape
self.BoundingBoxes.meta.shape = shape
self.BoundingBoxes.meta.dtype = np.object
self.BoundingBoxes.meta.axistags = vigra.defaultAxistags('txyzc')
def execute(self, slot, subindex, roi, result):
# check the axes - cannot do this in setupOutputs because we could be
# in some invalid intermediate state where the dimensions do not agree
shape = self.LabelImage.meta.shape
agree = [i == j for i, j in zip(self.Prediction.meta.shape, shape)]
assert all(agree),\
"shape mismatch: {} vs. {}".format(self.Prediction.meta.shape,
shape)
if slot == self.BoundingBoxes:
return self._execute_bbox(roi, result)
elif slot == self.Output:
self._execute_graphcut(roi, result)
else:
raise NotImplementedError(
"execute() is not implemented for slot {}".format(str(slot)))
def _execute_bbox(self, roi, result):
cc = self.LabelImage.get(roi).wait()
cc = vigra.taggedView(cc, axistags=self.LabelImage.meta.axistags)
cc = cc.withAxes(*'xyz')
logger.debug("computing bboxes...")
feats = vigra.analysis.extractRegionFeatures(
cc.astype(np.float32),
cc.astype(np.uint32),
features=["Count", "Coord<Minimum>", "Coord<Maximum>"])
feats_dict = {}
feats_dict["Coord<Minimum>"] = feats["Coord<Minimum>"]
feats_dict["Coord<Maximum>"] = feats["Coord<Maximum>"]
feats_dict["Count"] = feats["Count"]
return feats_dict
def _execute_graphcut(self, roi, result):
for i in (0, 4):
assert roi.stop[i] - roi.start[i] == 1,\
"Invalid roi for graph-cut: {}".format(str(roi))
t = roi.start[0]
c = roi.start[4]
margin = self.Margin.value
beta = self.Beta.value
MAXBOXSIZE = 10000000 # FIXME justification??
## request the bounding box coordinates ##
# the trailing index brackets give us the dictionary (instead of an
# array of size 1)
feats = self.BoundingBoxes.get(roi).wait()
mins = feats["Coord<Minimum>"]
maxs = feats["Coord<Maximum>"]
nobj = mins.shape[0]
# these are indices, so they should have an index datatype
mins = mins.astype(np.uint32)
maxs = maxs.astype(np.uint32)
## request the prediction image ##
pred = self.Prediction.get(roi).wait()
pred = vigra.taggedView(pred, axistags=self.Prediction.meta.axistags)
pred = pred.withAxes(*'xyz')
## request the connected components image ##
cc = self.LabelImage.get(roi).wait()
cc = vigra.taggedView(cc, axistags=self.LabelImage.meta.axistags)
cc = cc.withAxes(*'xyz')
# provide xyz view for the output (just need 8bit for segmentation
resultXYZ = vigra.taggedView(np.zeros(cc.shape, dtype=np.uint8),
axistags='xyz')
def processSingleObject(i):
logger.debug("processing object {}".format(i))
# maxs are inclusive, so we need to add 1
xmin = max(mins[i][0] - margin[0], 0)
ymin = max(mins[i][1] - margin[1], 0)
zmin = max(mins[i][2] - margin[2], 0)
xmax = min(maxs[i][0] + margin[0] + 1, cc.shape[0])
ymax = min(maxs[i][1] + margin[1] + 1, cc.shape[1])
zmax = min(maxs[i][2] + margin[2] + 1, cc.shape[2])
ccbox = cc[xmin:xmax, ymin:ymax, zmin:zmax]
resbox = resultXYZ[xmin:xmax, ymin:ymax, zmin:zmax]
nVoxels = ccbox.size
if nVoxels > MAXBOXSIZE:
#problem too large to run graph cut, assign to seed
logger.warn("Object {} too large for graph cut.".format(i))
resbox[ccbox == i] = 1
return
probbox = pred[xmin:xmax, ymin:ymax, zmin:zmax]
gcsegm = segmentGC(probbox, beta)
gcsegm = vigra.taggedView(gcsegm, axistags='xyz')
ccsegm = vigra.analysis.labelVolumeWithBackground(
gcsegm.astype(np.uint8))
# Extended bboxes of different objects might overlap.
# To avoid conflicting segmentations, we find all connected
# components in the results and only take the one, which
# overlaps with the object "core" or "seed", defined by the
# pre-thresholding
seed = ccbox == i
filtered = seed * ccsegm
passed = vigra.analysis.unique(filtered.astype(np.uint32))
assert len(passed.shape) == 1
if passed.size > 2:
logger.warn("ambiguous label assignment for region {}".format(
(xmin, xmax, ymin, ymax, zmin, zmax)))
resbox[ccbox == i] = 1
elif passed.size <= 1:
logger.warn("box {} segmented out with beta {}".format(
i, beta))
else:
# assign to the overlap region
label = passed[1] # 0 is background
resbox[ccsegm == label] = 1
pool = RequestPool()
#FIXME make sure that the parallel computations fit into memory
for i in range(1, nobj):
req = Request(functools.partial(processSingleObject, i))
pool.add(req)
logger.info("Processing {} objects ...".format(nobj - 1))
pool.wait()
pool.clean()
logger.info("object loop done")
# prepare result
resView = vigra.taggedView(result, axistags=self.Output.meta.axistags)
resView = resView.withAxes(*'xyz')
# some labels could have been removed => relabel
vigra.analysis.labelVolumeWithBackground(resultXYZ, out=resView)
def propagateDirty(self, slot, subindex, roi):
super(OpObjectsSegment, self).propagateDirty(slot, subindex, roi)
if slot == self.LabelImage:
# time-channel slices are pairwise independent
# determine t, c from input volume
t_ind = 0
c_ind = 4
t = (roi.start[t_ind], roi.stop[t_ind])
c = (roi.start[c_ind], roi.stop[c_ind])
# set output dirty
start = t[0:1] + (0, ) * 3 + c[0:1]
stop = t[1:2] + self.Output.meta.shape[1:4] + c[1:2]
roi = SubRegion(self.Output, start=start, stop=stop)
self.Output.setDirty(roi)
elif slot == self.Margin:
# margin affects the whole volume
self.Output.setDirty(slice(None))
class OpGraphCut(Operator):
name = "OpGraphCut"
# prediction maps
Prediction = InputSlot()
# graph cut parameter, usually called lambda
Beta = InputSlot(value=.2)
# labeled segmentation image
# i=0: background
# i>0: connected foreground object i
Output = OutputSlot()
CachedOutput = OutputSlot()
def __init__(self, *args, **kwargs):
super(OpGraphCut, self).__init__(*args, **kwargs)
self._cache = None
def setupOutputs(self):
# sanity checks
shape = self.Prediction.meta.shape
assert len(shape) == 5,\
"Prediction maps must be a full 5d volume (tzyxc)"
tags = self.Prediction.meta.getAxisKeys()
tags = "".join(tags)
assert tags == 'tzyxc',\
"Prediction maps have wrong axes order"\
"(expected: tzyxc, got: {})".format(tags)
if self._cache is not None:
self.CachedOutput.disconnect()
self._cache.cleanUp()
self._cache = None
cache = OpCompressedCache(parent=self)
cache.name = "{}._cache".format(self.name)
cache.Input.connect(self.Output)
self._cache = cache
self.CachedOutput.connect(self._cache.Output)
self.Output.meta.assignFrom(self.Prediction.meta)
# output is a label image
self.Output.meta.dtype = np.uint32
# cache should hold entire c-t-slices in memory
shape = list(self.Prediction.meta.shape)
shape[0] = 1
shape[4] = 1
self._cache.BlockShape.setValue(tuple(shape))
def execute(self, slot, subindex, roi, result):
assert slot == self.Output, "Unknown slot requested: {}".format(slot)
for i in (0, 4):
assert roi.stop[i] - roi.start[i] == 1,\
"Invalid roi for graph-cut: {}".format(str(roi))
## request the prediction image ##
pred = self.Prediction.get(roi).wait()
pred = vigra.taggedView(pred, axistags=self.Prediction.meta.axistags)
pred = pred.withAxes(*'zyx')
# prepare result
resView = vigra.taggedView(result, axistags=self.Output.meta.axistags)
resView = resView.withAxes(*'zyx')
logger.info("Executing graph cut ... (this might take a while)")
threshold_binary = segmentGC(pred, self.Beta.value)
threshold_binary = vigra.taggedView( threshold_binary, 'zyx' )
logger.info("Graph-cut done")
# label the segmentation so that this operator is consistent with
# the other thresholding operators
vigra.analysis.labelVolumeWithBackground(threshold_binary.astype(np.uint8), out=resView)
def propagateDirty(self, slot, subindex, roi):
# all input slots affect the (global) graph cut computation
if slot == self.Beta:
# beta value affects the whole volume
self.Output.setDirty(slice(None))
elif slot == self.Prediction:
# time-channel slices are pairwise independent
# determine t, c from input volume
t_ind = 0
c_ind = 4
t = (roi.start[t_ind], roi.stop[t_ind])
c = (roi.start[c_ind], roi.stop[c_ind])
# set output dirty
start = t[0:1] + (0,)*3 + c[0:1]
stop = t[1:2] + self.Output.meta.shape[1:4] + c[1:2]
roi = SubRegion(self.Output, start=start, stop=stop)
self.Output.setDirty(roi)
class OpLabelingABC(Operator):
__metaclass__ = ABCMeta
## input with axes 'xyzct'
Input = InputSlot()
## background with axes 'xyzct', spatial axes must be singletons
Background = InputSlot()
Output = OutputSlot()
CachedOutput = OutputSlot()
# the numeric type that is used for labeling
labelType = np.uint32
## list of supported dtypes
@abstractproperty
def supportedDtypes(self):
pass
def __init__(self, *args, **kwargs):
super(OpLabelingABC, self).__init__(*args, **kwargs)
self._cache = OpCompressedCache(parent=self)
self._cache.name = "OpLabelVolume.OutputCache"
self._cache.Input.connect(self.Output)
self.CachedOutput.connect(self._cache.Output)
def setupOutputs(self):
# check if the input dtype is valid
if self.Input.ready():
dtype = self.Input.meta.dtype
if dtype not in self.supportedDtypes:
msg = "{}: dtype '{}' not supported "\
"with method 'vigra'. Supported types: {}"
msg = msg.format(self.name, dtype, self.supportedDtypes)
raise ValueError(msg)
# set cache chunk shape to the whole spatial volume
shape = np.asarray(self.Input.meta.shape, dtype=np.int)
shape[3:5] = 1
self._cache.BlockShape.setValue(tuple(shape))
# setup meta for Output
self.Output.meta.assignFrom(self.Input.meta)
self.Output.meta.dtype = self.labelType
def propagateDirty(self, slot, subindex, roi):
# a change in either input or background makes the whole
# time-channel-slice dirty (CCL is a global operation)
outroi = roi.copy()
outroi.start[:3] = (0, 0, 0)
outroi.stop[:3] = self.Input.meta.shape[:3]
self.Output.setDirty(outroi)
self.CachedOutput.setDirty(outroi)
def execute(self, slot, subindex, roi, result):
if slot == self.Output:
# just label the ROI and write it to result
self._label(roi, result)
else:
raise ValueError("Request to unknown slot {}".format(slot))
def _label(self, roi, result):
result = vigra.taggedView(result, axistags=self.Output.meta.axistags)
# get the background values
bg = self.Background[...].wait()
bg = vigra.taggedView(bg, axistags=self.Background.meta.axistags)
bg = bg.withAxes(*'ct')
assert np.all(self.Background.meta.shape[3:] ==
self.Input.meta.shape[3:]),\
"Shape of background values incompatible to shape of Input"
# do labeling in parallel over channels and time slices
pool = RequestPool()
start = np.asarray(roi.start, dtype=np.int)
stop = np.asarray(roi.stop, dtype=np.int)
for ti, t in enumerate(range(roi.start[4], roi.stop[4])):
start[4], stop[4] = t, t + 1
for ci, c in enumerate(range(roi.start[3], roi.stop[3])):
start[3], stop[3] = c, c + 1
newRoi = SubRegion(self.Output,
start=tuple(start),
stop=tuple(stop))
resView = result[..., ci, ti].withAxes(*'xyz')
req = Request(partial(self._label3d, newRoi, bg[c, t],
resView))
pool.add(req)
logger.debug(
"{}: Computing connected components for ROI {} ...".format(
self.name, roi))
pool.wait()
pool.clean()
logger.debug("{}: Connected components computed.".format(self.name))
## compute the requested roi and put the results into result
#
# @param result the array to write into, 3d xyz
@abstractmethod
def _label3d(self, roi, bg, result):
pass