def execute(self, slot, subindex, roi, result):
t1 = time.perf_counter()
key = roi.toSlice()
nlabels = self.inputs["LabelsCount"].value
traceLogger.debug(
"OpPredictRandomForest: Requesting classifier. roi={}".format(roi))
forests = self.inputs["Classifier"][:].wait()
if any(forest is None for forest in forests):
# Training operator may return 'None' if there was no data to train with
return np.zeros(np.subtract(roi.stop, roi.start),
dtype=np.float32)[...]
traceLogger.debug("OpPredictRandomForest: Got classifier")
#assert RF.labelCount() == nlabels, "ERROR: OpPredictRandomForest, labelCount differs from true labelCount! %r vs. %r" % (RF.labelCount(), nlabels)
newKey = key[:-1]
newKey += (slice(0, self.inputs["Image"].meta.shape[-1], None), )
res = self.inputs["Image"][newKey].wait()
shape = res.shape
prod = np.prod(shape[:-1])
res.shape = (prod, shape[-1])
features = res
predictions = [0] * len(forests)
t2 = time.perf_counter()
pool = RequestPool()
def predict_forest(i):
predictions[i] = forests[i].predict(
np.asarray(features, dtype=np.float32))
predictions[i] = predictions[i].reshape(result.shape[:-1])
for i, f in enumerate(forests):
req = pool.request(partial(predict_forest, i))
pool.wait()
pool.clean()
#predictions[0] = forests[0].predict(np.asarray(features, dtype = np.float32), normalize = False)
#predictions[0] = predictions[0].reshape(result.shape)
prediction = np.dstack(predictions)
result[...] = prediction
# If our LabelsCount is higher than the number of labels in the training set,
# then our results aren't really valid. FIXME !!!
# Duplicate the last label's predictions
#for c in range(result.shape[-1]):
# result[...,c] = prediction[...,min(c+key[-1].start, prediction.shape[-1]-1)]
t3 = time.perf_counter()
logger.debug(
"Predict took %fseconds, actual RF time was %fs, feature time was %fs"
% (t3 - t1, t3 - t2, t2 - t1))
return result
def execute(self, slot, subindex, roi, result):
assert slot == self._ReorderedOutput
pool = RequestPool()
t_ind = 0
for t in range(roi.start[0], roi.stop[0]):
c_ind = 0
for c in range(roi.start[-1], roi.stop[-1]):
newroi = roi.copy()
newroi.start[0] = t
newroi.stop[0] = t+1
newroi.start[-1] = c
newroi.stop[-1] = c+1
req = self._op.Output.get(newroi)
resView = result[t_ind:t_ind+1, ..., c_ind:c_ind+1]
req.writeInto(resView)
pool.add(req)
c_ind += 1
t_ind += 1
pool.wait()
pool.clean()
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))
def execute(self, slot, subindex, roi, result):
with self._lock:
if self.cache is None:
fullBlockShape = numpy.array([self.blockShape.value for i in self.Input.meta.shape])
fun = self.inputs["Function"].value
#data = self.inputs["Input"][:].wait()
#split up requests into blocks
shape = self.Input.meta.shape
numBlocks = numpy.ceil(shape/(1.0*fullBlockShape)).astype("int")
blockCache = numpy.ndarray(shape = numpy.prod(numBlocks), dtype=self.Output.meta.dtype)
pool = RequestPool()
#blocks holds the different roi keys for each of the blocks
blocks = itertools.product(*[range(i) for i in numBlocks])
blockKeys = []
for b in blocks:
start = b * fullBlockShape
stop = b * fullBlockShape + fullBlockShape
stop = numpy.min(numpy.vstack((stop, shape)), axis=0)
blockKey = roiToSlice(start, stop)
blockKeys.append(blockKey)
def predict_block(i):
data = self.Input[blockKeys[i]].wait()
blockCache[i] = fun(data)
for i,f in enumerate(blockCache):
req = pool.request(partial(predict_block,i))
pool.wait()
pool.clean()
self.cache = [fun(blockCache)]
return self.cache
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))
def execute(self, slot, subindex, rroi, result):
key = roiToSlice(rroi.start,rroi.stop)
cnt = 0
written = 0
start, stop = roi.sliceToRoi(key, self.outputs["Output"].meta.shape)
assert (stop<=self.outputs["Output"].meta.shape).all()
#axisindex = self.inputs["AxisIndex"].value
flag = self.inputs["AxisFlag"].value
axisindex = self.outputs["Output"].meta.axistags.index(flag)
#ugly-ugly-ugly
oldkey = list(key)
oldkey.pop(axisindex)
#print "STACKER: ", flag, axisindex
#print "requesting an outslot from stacker:", key, result.shape
#print "input slots total: ", len(self.inputs['Images'])
requests = []
pool = RequestPool()
for i, inSlot in enumerate(self.inputs['Images']):
req = None
inTagKeys = [ax.key for ax in inSlot.meta.axistags]
if flag in inTagKeys:
slices = inSlot.meta.shape[axisindex]
if cnt + slices >= start[axisindex] and start[axisindex]-cnt<slices and start[axisindex]+written<stop[axisindex]:
begin = 0
if cnt < start[axisindex]:
begin = start[axisindex] - cnt
end = slices
if cnt + end > stop[axisindex]:
end -= cnt + end - stop[axisindex]
key_ = copy.copy(oldkey)
key_.insert(axisindex, slice(begin, end, None))
reskey = [slice(None, None, None) for x in range(len(result.shape))]
reskey[axisindex] = slice(written, written+end-begin, None)
req = inSlot[tuple(key_)].writeInto(result[tuple(reskey)])
written += end - begin
cnt += slices
else:
if cnt>=start[axisindex] and start[axisindex] + written < stop[axisindex]:
#print "key: ", key, "reskey: ", reskey, "oldkey: ", oldkey
#print "result: ", result.shape, "inslot:", inSlot.meta.shape
reskey = [slice(None, None, None) for s in oldkey]
reskey.insert(axisindex, written)
destArea = result[tuple(reskey)]
req = inSlot[tuple(oldkey)].writeInto(destArea)
written += 1
cnt += 1
if req is not None:
pool.add(req)
pool.wait()
pool.clean()
def execute(self, slot, subindex, rroi, result):
key = roiToSlice(rroi.start,rroi.stop)
cnt = 0
written = 0
start, stop = roi.sliceToRoi(key, self.outputs["Output"].meta.shape)
assert (stop<=self.outputs["Output"].meta.shape).all()
#axisindex = self.inputs["AxisIndex"].value
flag = self.inputs["AxisFlag"].value
axisindex = self.outputs["Output"].meta.axistags.index(flag)
#ugly-ugly-ugly
oldkey = list(key)
oldkey.pop(axisindex)
#print "STACKER: ", flag, axisindex
#print "requesting an outslot from stacker:", key, result.shape
#print "input slots total: ", len(self.inputs['Images'])
requests = []
pool = RequestPool()
for i, inSlot in enumerate(self.inputs['Images']):
req = None
inTagKeys = [ax.key for ax in inSlot.meta.axistags]
if flag in inTagKeys:
slices = inSlot.meta.shape[axisindex]
if cnt + slices >= start[axisindex] and start[axisindex]-cnt<slices and start[axisindex]+written<stop[axisindex]:
begin = 0
if cnt < start[axisindex]:
begin = start[axisindex] - cnt
end = slices
if cnt + end > stop[axisindex]:
end -= cnt + end - stop[axisindex]
key_ = copy.copy(oldkey)
key_.insert(axisindex, slice(begin, end, None))
reskey = [slice(None, None, None) for x in range(len(result.shape))]
reskey[axisindex] = slice(written, written+end-begin, None)
req = inSlot[tuple(key_)].writeInto(result[tuple(reskey)])
written += end - begin
cnt += slices
else:
if cnt>=start[axisindex] and start[axisindex] + written < stop[axisindex]:
#print "key: ", key, "reskey: ", reskey, "oldkey: ", oldkey
#print "result: ", result.shape, "inslot:", inSlot.meta.shape
reskey = [slice(None, None, None) for s in oldkey]
reskey.insert(axisindex, written)
destArea = result[tuple(reskey)]
req = inSlot[tuple(oldkey)].writeInto(destArea)
written += 1
cnt += 1
if req is not None:
pool.add(req)
pool.wait()
pool.clean()
def execute(self, slot, subindex, roi, result):
t1 = time.time()
key = roi.toSlice()
nlabels=self.inputs["LabelsCount"].value
traceLogger.debug("OpPredictRandomForest: Requesting classifier. roi={}".format(roi))
forests=self.inputs["Classifier"][:].wait()
if forests is None or any(x is None for x in forests):
# Training operator may return 'None' if there was no data to train with
return numpy.zeros(numpy.subtract(roi.stop, roi.start), dtype=numpy.float32)[...]
traceLogger.debug("OpPredictRandomForest: Got classifier")
#assert RF.labelCount() == nlabels, "ERROR: OpPredictRandomForest, labelCount differs from true labelCount! %r vs. %r" % (RF.labelCount(), nlabels)
newKey = key[:-1]
newKey += (slice(0,self.inputs["Image"].meta.shape[-1],None),)
res = self.inputs["Image"][newKey].wait()
shape=res.shape
prod = numpy.prod(shape[:-1])
res.shape = (prod, shape[-1])
features=res
predictions = [0]*len(forests)
def predict_forest(number):
predictions[number] = forests[number].predictProbabilities(numpy.asarray(features, dtype=numpy.float32))
t2 = time.time()
# predict the data with all the forests in parallel
pool = RequestPool()
for i,f in enumerate(forests):
req = pool.request(partial(predict_forest, i))
pool.wait()
pool.clean()
prediction=numpy.dstack(predictions)
prediction = numpy.average(prediction, axis=2)
prediction.shape = shape[:-1] + (forests[0].labelCount(),)
#prediction = prediction.reshape(*(shape[:-1] + (forests[0].labelCount(),)))
# If our LabelsCount is higher than the number of labels in the training set,
# then our results aren't really valid. FIXME !!!
# Duplicate the last label's predictions
for c in range(result.shape[-1]):
result[...,c] = prediction[...,min(c+key[-1].start, prediction.shape[-1]-1)]
t3 = time.time()
self.logger.debug("predict roi=%r took %fseconds, actual RF time was %fs, feature time was %fs" % (key, t3-t1, t3-t2, t2-t1))
return result
def execute(self, slot, subindex, roi, result):
featList = []
labelsList = []
for i in range(len(self.Labels)):
feats = self.Features[i]([]).wait()
# TODO: we should be able to use self.Labels[i].value,
# but the current implementation of Slot.value() does not
# do the right thing.
labels = self.Labels[i]([]).wait()
featstmp, labelstmp = make_feature_array(feats, labels)
featList.append(featstmp)
labelsList.append(labelstmp)
featMatrix = _concatenate(featList, axis=0)
labelsMatrix = _concatenate(labelsList, axis=0)
print "training on matrix:", featMatrix.shape, featMatrix.dtype
if len(featMatrix) == 0 or len(labelsMatrix) == 0:
result[:] = None
return
oob = [0] * self.ForestCount.value
try:
# Ensure there are no NaNs in the feature matrix
# TODO: There should probably be a better way to fix this...
featMatrix = numpy.asarray(featMatrix, dtype=numpy.float32)
nanFeatMatrix = numpy.isnan(featMatrix)
if nanFeatMatrix.any():
warnings.warn("Feature matrix has NaN values! Replacing with 0.0...")
featMatrix[numpy.where(nanFeatMatrix)] = 0.0
# train and store forests in parallel
pool = RequestPool()
for i in range(self.ForestCount.value):
def train_and_store(number):
result[number] = vigra.learning.RandomForest(self._tree_count)
oob[number] = result[number].learnRF(featMatrix, numpy.asarray(labelsMatrix, dtype=numpy.uint32))
print "intermediate oob:", oob[number]
req = Request( partial(train_and_store, i) )
pool.add( req )
pool.wait()
pool.clean()
except:
print ("couldn't learn classifier")
raise
oob_total = numpy.mean(oob)
print "training finished, out of bag error:", oob_total
return result
def predict(cls, X, method="classic"):
"""
predict if the histograms in X correspond to missing regions
do this for subsets of X in parallel
"""
if cls._manager is None:
cls._manager = SVMManager()
assert len(
X.shape
) == 2, "Prediction data must have shape (nSamples, nHistogramBins)."
nBins = X.shape[1]
if method == "classic":
svm = PseudoSVC()
else:
try:
svm = cls._manager.get(nBins)
except SVMManager.NotTrainedError:
# fail gracefully if not trained => responsibility of user!
svm = PseudoSVC()
y = np.zeros((len(X), )) * np.nan
pool = RequestPool()
chunkSize = 1000 # FIXME magic number??
nChunks = len(X) // chunkSize + (1 if len(X) % chunkSize > 0 else 0)
s = [
slice(k * chunkSize, min((k + 1) * chunkSize, len(X)))
for k in range(nChunks)
]
def partFun(i):
y[s[i]] = svm.predict(X[s[i]])
for i in range(nChunks):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
# not neccessary
# assert not np.any(np.isnan(y))
return np.asarray(y)
def predict(cls, X, method='classic'):
"""
predict if the histograms in X correspond to missing regions
do this for subsets of X in parallel
"""
if cls._manager is None:
cls._manager = SVMManager()
# svm input has to be (nSamples, nFeatures) -> for us: (nSampels = len(X), nFeatures = # of histogrambins )
X_reshaped = np.zeros((len(X), len(X[0])))
for i in range(len(X)):
X_reshaped[i, :] = X[i]
n_bins = len(X[0])
if method == 'classic' or not have_sklearn:
logger.warning("no real svm used! -> PseudoSVC")
svm = PseudoSVC()
else:
# load samples for histograms of labeled regions
try:
svm = cls._manager.get(n_bins)
except SVMManager.NotTrainedError:
# fail gracefully if not trained => responsibility of user!
svm = PseudoSVC()
y = np.zeros((len(X),)) * np.nan
pool = RequestPool()
# chunk up all samples from X into chunks that will be predicted in parallel
chunk_size = 1000 # FIXME magic number??
n_chunks = len(X)/chunk_size + (1 if len(X) % chunk_size > 0 else 0)
s = [slice(k * chunk_size, min((k + 1) * chunk_size, len(X)))
for k in range(n_chunks)]
def partFun(i):
y[s[i]] = svm.predict(X_reshaped[s[i], :])
for i in range(n_chunks):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
return np.asarray(y)
def execute(self, slot, subindex, roi, result):
with self._lock:
if self.cache is None:
shape = self.Input.meta.shape
# self.blockshape has None in the last dimension to indicate that it should not be
# handled block-wise. None is replaced with the image shape in the respective axis.
fullBlockShape = []
for u, v in zip(self.blockShape.value, shape):
if u is not None:
fullBlockShape.append(u)
else:
fullBlockShape.append(v)
fullBlockShape = numpy.array(fullBlockShape,
dtype=numpy.float64)
# data = self.inputs["Input"][:].wait()
# split up requests into blocks
numBlocks = numpy.ceil(shape / fullBlockShape).astype("int")
blockCache = numpy.ndarray(shape=numpy.prod(numBlocks),
dtype=self.Output.meta.dtype)
pool = RequestPool()
# blocks holds the different roi keys for each of the blocks
blocks = itertools.product(
*[list(range(i)) for i in numBlocks])
blockKeys = []
for b in blocks:
start = b * fullBlockShape
stop = b * fullBlockShape + fullBlockShape
stop = numpy.min(numpy.vstack((stop, shape)), axis=0)
blockKey = roiToSlice(start, stop)
blockKeys.append(blockKey)
fun = self.inputs["Function"].value
def predict_block(i):
data = self.Input[blockKeys[i]].wait()
blockCache[i] = fun(data)
for i, f in enumerate(blockCache):
req = pool.request(partial(predict_block, i))
pool.wait()
pool.clean()
self.cache = [fun(blockCache)]
return self.cache
def predict(cls, X, method='classic'):
'''
predict if the histograms in X correspond to missing regions
do this for subsets of X in parallel
'''
if cls._manager is None:
cls._manager = SVMManager()
assert len(X.shape) == 2, \
"Prediction data must have shape (nSamples, nHistogramBins)."
nBins = X.shape[1]
if method == 'classic' or not havesklearn:
svm = PseudoSVC()
else:
try:
svm = cls._manager.get(nBins)
except SVMManager.NotTrainedError:
# fail gracefully if not trained => responsibility of user!
svm = PseudoSVC()
y = np.zeros((len(X),))*np.nan
pool = RequestPool()
chunkSize = 1000 # FIXME magic number??
nChunks = len(X)//chunkSize + (1 if len(X) % chunkSize > 0 else 0)
s = [slice(k*chunkSize, min((k+1)*chunkSize, len(X)))
for k in range(nChunks)]
def partFun(i):
y[s[i]] = svm.predict(X[s[i]])
for i in range(nChunks):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
# not neccessary
#assert not np.any(np.isnan(y))
return np.asarray(y)
def execute(self, slot, subindex, roi, result):
with self._lock:
if self.cache is None:
shape = self.Input.meta.shape
# self.blockshape has None in the last dimension to indicate that it should not be
# handled block-wise. None is replaced with the image shape in the respective axis.
fullBlockShape = []
for u, v in zip(self.blockShape.value, shape):
if u is not None:
fullBlockShape.append(u)
else:
fullBlockShape.append(v)
fullBlockShape = numpy.array(fullBlockShape, dtype=numpy.float64)
#data = self.inputs["Input"][:].wait()
#split up requests into blocks
numBlocks = numpy.ceil(shape / fullBlockShape).astype("int")
blockCache = numpy.ndarray(shape = numpy.prod(numBlocks), dtype=self.Output.meta.dtype)
pool = RequestPool()
#blocks holds the different roi keys for each of the blocks
blocks = itertools.product(*[list(range(i)) for i in numBlocks])
blockKeys = []
for b in blocks:
start = b * fullBlockShape
stop = b * fullBlockShape + fullBlockShape
stop = numpy.min(numpy.vstack((stop, shape)), axis=0)
blockKey = roiToSlice(start, stop)
blockKeys.append(blockKey)
fun = self.inputs["Function"].value
def predict_block(i):
data = self.Input[blockKeys[i]].wait()
blockCache[i] = fun(data)
for i,f in enumerate(blockCache):
req = pool.request(partial(predict_block,i))
pool.wait()
pool.clean()
self.cache = [fun(blockCache)]
return self.cache
def execute(self, slot, subindex, roi, result):
with self._lock:
if self.cache is None:
fullBlockShape = numpy.array(
[self.blockShape.value for i in self.Input.meta.shape])
fun = self.inputs["Function"].value
#data = self.inputs["Input"][:].wait()
#split up requests into blocks
shape = self.Input.meta.shape
numBlocks = numpy.ceil(shape /
(1.0 * fullBlockShape)).astype("int")
blockCache = numpy.ndarray(shape=numpy.prod(numBlocks),
dtype=self.Output.meta.dtype)
pool = RequestPool()
#blocks holds the different roi keys for each of the blocks
blocks = itertools.product(*[range(i) for i in numBlocks])
blockKeys = []
for b in blocks:
start = b * fullBlockShape
stop = b * fullBlockShape + fullBlockShape
stop = numpy.min(numpy.vstack((stop, shape)), axis=0)
blockKey = roiToSlice(start, stop)
blockKeys.append(blockKey)
def predict_block(i):
data = self.Input[blockKeys[i]].wait()
blockCache[i] = fun(data)
for i, f in enumerate(blockCache):
req = pool.request(partial(predict_block, i))
pool.wait()
pool.clean()
self.cache = [fun(blockCache)]
return self.cache
def execute(self, slot, subindex, rroi, result):
assert slot == self.Features or slot == self.Output
if slot == self.Features:
key = roiToSlice(rroi.start, rroi.stop)
index = subindex[0]
key = list(key)
channelIndex = self.Input.meta.axistags.index('c')
# Translate channel slice to the correct location for the output slot.
key[channelIndex] = slice(self.featureOutputChannels[index][0] + key[channelIndex].start,
self.featureOutputChannels[index][0] + key[channelIndex].stop)
rroi = SubRegion(self.Output, pslice=key)
# Get output slot region for this channel
return self.execute(self.Output, (), rroi, result)
elif slot == self.outputs["Output"]:
key = rroi.toSlice()
logger.debug("OpPixelFeaturesPresmoothed: request %s" % (rroi.pprint(),))
cnt = 0
written = 0
assert (rroi.stop<=self.outputs["Output"].meta.shape).all()
flag = 'c'
channelAxis=self.inputs["Input"].meta.axistags.index('c')
axisindex = channelAxis
oldkey = list(key)
oldkey.pop(axisindex)
inShape = self.inputs["Input"].meta.shape
hasChannelAxis = (self.Input.meta.axistags.axisTypeCount(vigra.AxisType.Channels) > 0)
#if (self.Input.meta.axistags.axisTypeCount(vigra.AxisType.Channels) == 0):
# noChannels = True
inAxistags = self.inputs["Input"].meta.axistags
shape = self.outputs["Output"].meta.shape
axistags = self.outputs["Output"].meta.axistags
result = result.view(vigra.VigraArray)
result.axistags = copy.copy(axistags)
hasTimeAxis = self.inputs["Input"].meta.axistags.axisTypeCount(vigra.AxisType.Time)
timeAxis=self.inputs["Input"].meta.axistags.index('t')
subkey = popFlagsFromTheKey(key,axistags,'c')
subshape=popFlagsFromTheKey(shape,axistags,'c')
at2 = copy.copy(axistags)
at2.dropChannelAxis()
subshape=popFlagsFromTheKey(subshape,at2,'t')
subkey = popFlagsFromTheKey(subkey,at2,'t')
oldstart, oldstop = roi.sliceToRoi(key, shape)
start, stop = roi.sliceToRoi(subkey,subkey)
maxSigma = max(0.7,self.maxSigma) #we use 0.7 as an approximation of not doing any smoothing
#smoothing was already applied previously
# The region of the smoothed image we need to give to the feature filter (in terms of INPUT coordinates)
# 0.7, because the features receive a pre-smoothed array and don't need much of a neighborhood
vigOpSourceStart, vigOpSourceStop = roi.enlargeRoiForHalo(start, stop, subshape, 0.7, self.WINDOW_SIZE)
# The region of the input that we need to give to the smoothing operator (in terms of INPUT coordinates)
newStart, newStop = roi.enlargeRoiForHalo(vigOpSourceStart, vigOpSourceStop, subshape, maxSigma, self.WINDOW_SIZE)
newStartSmoother = roi.TinyVector(start - vigOpSourceStart)
newStopSmoother = roi.TinyVector(stop - vigOpSourceStart)
roiSmoother = roi.roiToSlice(newStartSmoother, newStopSmoother)
# Translate coordinates (now in terms of smoothed image coordinates)
vigOpSourceStart = roi.TinyVector(vigOpSourceStart - newStart)
vigOpSourceStop = roi.TinyVector(vigOpSourceStop - newStart)
readKey = roi.roiToSlice(newStart, newStop)
writeNewStart = start - newStart
writeNewStop = writeNewStart + stop - start
treadKey=list(readKey)
if hasTimeAxis:
if timeAxis < channelAxis:
treadKey.insert(timeAxis, key[timeAxis])
else:
treadKey.insert(timeAxis-1, key[timeAxis])
if self.inputs["Input"].meta.axistags.axisTypeCount(vigra.AxisType.Channels) == 0:
treadKey = popFlagsFromTheKey(treadKey,axistags,'c')
else:
treadKey.insert(channelAxis, slice(None,None,None))
treadKey=tuple(treadKey)
req = self.inputs["Input"][treadKey]
sourceArray = req.wait()
req.clean()
#req.result = None
req.destination = None
if sourceArray.dtype != numpy.float32:
sourceArrayF = sourceArray.astype(numpy.float32)
try:
sourceArray.resize((1,), refcheck = False)
except:
pass
del sourceArray
sourceArray = sourceArrayF
#if (self.Input.meta.axistags.axisTypeCount(vigra.AxisType.Channels) == 0):
#add a channel dimension to make the code afterwards more uniform
# sourceArray = sourceArray.view(numpy.ndarray)
# sourceArray = sourceArray.reshape(sourceArray.shape+(1,))
sourceArrayV = sourceArray.view(vigra.VigraArray)
sourceArrayV.axistags = copy.copy(inAxistags)
dimCol = len(self.scales)
dimRow = self.matrix.shape[0]
sourceArraysForSigmas = [None]*dimCol
#connect individual operators
try:
for j in range(dimCol):
hasScale = False
for i in range(dimRow):
if self.matrix[i,j]:
hasScale = True
if not hasScale:
continue
destSigma = 1.0
if self.scales[j] > destSigma:
tempSigma = math.sqrt(self.scales[j]**2 - destSigma**2)
else:
destSigma = 0.0
tempSigma = self.scales[j]
vigOpSourceShape = list(vigOpSourceStop - vigOpSourceStart)
if hasTimeAxis:
if timeAxis < channelAxis:
vigOpSourceShape.insert(timeAxis, ( oldstop - oldstart)[timeAxis])
else:
vigOpSourceShape.insert(timeAxis-1, ( oldstop - oldstart)[timeAxis])
vigOpSourceShape.insert(channelAxis, inShape[channelAxis])
sourceArraysForSigmas[j] = numpy.ndarray(tuple(vigOpSourceShape),numpy.float32)
for i,vsa in enumerate(sourceArrayV.timeIter()):
droi = (tuple(vigOpSourceStart._asint()), tuple(vigOpSourceStop._asint()))
tmp_key = getAllExceptAxis(len(sourceArraysForSigmas[j].shape),timeAxis, i)
sourceArraysForSigmas[j][tmp_key] = self._computeGaussianSmoothing(vsa, tempSigma, droi)
else:
droi = (tuple(vigOpSourceStart._asint()), tuple(vigOpSourceStop._asint()))
sourceArraysForSigmas[j] = self._computeGaussianSmoothing(sourceArrayV, tempSigma, droi)
except RuntimeError as e:
if e.message.find('kernel longer than line') > -1:
message = "Feature computation error:\nYour image is too small to apply a filter with sigma=%.1f. Please select features with smaller sigmas." % self.scales[j]
raise RuntimeError(message)
else:
raise e
del sourceArrayV
try:
sourceArray.resize((1,), refcheck = False)
except ValueError:
# Sometimes this fails, but that's okay.
logger.debug("Failed to free array memory.")
del sourceArray
closures = []
#connect individual operators
for i in range(dimRow):
for j in range(dimCol):
val=self.matrix[i,j]
if val:
vop= self.featureOps[i][j]
oslot = vop.outputs["Output"]
req = None
#inTagKeys = [ax.key for ax in oslot.meta.axistags]
#print inTagKeys, flag
if hasChannelAxis:
slices = oslot.meta.shape[axisindex]
if cnt + slices >= rroi.start[axisindex] and rroi.start[axisindex]-cnt<slices and rroi.start[axisindex]+written<rroi.stop[axisindex]:
begin = 0
if cnt < rroi.start[axisindex]:
begin = rroi.start[axisindex] - cnt
end = slices
if cnt + end > rroi.stop[axisindex]:
end -= cnt + end - rroi.stop[axisindex]
key_ = copy.copy(oldkey)
key_.insert(axisindex, slice(begin, end, None))
reskey = [slice(None, None, None) for x in range(len(result.shape))]
reskey[axisindex] = slice(written, written+end-begin, None)
destArea = result[tuple(reskey)]
#readjust the roi for the new source array
roiSmootherList = list(roiSmoother)
roiSmootherList.insert(axisindex, slice(begin, end, None))
if hasTimeAxis:
# The time slice in the ROI doesn't matter:
# The sourceArrayParameter below overrides the input data to be used.
roiSmootherList.insert(timeAxis, 0)
roiSmootherRegion = SubRegion(oslot, pslice=roiSmootherList)
closure = partial(oslot.operator.execute, oslot, (), roiSmootherRegion, destArea, sourceArray = sourceArraysForSigmas[j])
closures.append(closure)
written += end - begin
cnt += slices
else:
if cnt>=rroi.start[axisindex] and rroi.start[axisindex] + written < rroi.stop[axisindex]:
reskey = [slice(None, None, None) for x in range(len(result.shape))]
slices = oslot.meta.shape[axisindex]
reskey[axisindex]=slice(written, written+slices, None)
#print "key: ", key, "reskey: ", reskey, "oldkey: ", oldkey, "resshape:", result.shape
#print "roiSmoother:", roiSmoother
destArea = result[tuple(reskey)]
#print "destination area:", destArea.shape
logger.debug(oldkey, destArea.shape, sourceArraysForSigmas[j].shape)
oldroi = SubRegion(oslot, pslice=oldkey)
#print "passing roi:", oldroi
closure = partial(oslot.operator.execute, oslot, (), oldroi, destArea, sourceArray = sourceArraysForSigmas[j])
closures.append(closure)
written += 1
cnt += 1
pool = RequestPool()
for c in closures:
r = pool.request(c)
pool.wait()
pool.clean()
for i in range(len(sourceArraysForSigmas)):
if sourceArraysForSigmas[i] is not None:
try:
sourceArraysForSigmas[i].resize((1,))
except:
sourceArraysForSigmas[i] = None
def execute(self, slot, subindex, roi, result):
progress = 0
numImages = len(self.Images)
self.progressSignal(progress)
featMatrix=[]
labelsMatrix=[]
tagList = []
#result[0] = self._svr
for i,labels in enumerate(self.inputs["ForegroundLabels"]):
if labels.meta.shape is not None:
opGaussian = OpGaussianSmoothing(parent = self, graph = self.graph)
opGaussian.Sigma.setValue(self.Sigma.value)
opGaussian.Input.connect(self.ForegroundLabels[i])
blocks = self.inputs["nonzeroLabelBlocks"][i][0].wait()
reqlistlabels = []
reqlistbg = []
reqlistfeat = []
progress += 10 / numImages
self.progressSignal(progress)
for b in blocks[0]:
request = opGaussian.Output[b]
#request = labels[b]
featurekey = list(b)
featurekey[-1] = slice(None, None, None)
request2 = self.Images[i][featurekey]
request3 = self.inputs["BackgroundLabels"][i][b]
reqlistlabels.append(request)
reqlistfeat.append(request2)
reqlistbg.append(request3)
traceLogger.debug("Requests prepared")
numLabelBlocks = len(reqlistlabels)
progress_outer = [progress]
if numLabelBlocks > 0:
progressInc = (80 - 10)/(numLabelBlocks * numImages)
def progressNotify(req):
progress_outer[0] += progressInc/2
self.progressSignal(progress_outer[0])
for ir, req in enumerate(reqlistfeat):
req.notify_finished(progressNotify)
req.submit()
for ir, req in enumerate(reqlistlabels):
req.notify_finished(progressNotify)
req.submit()
for ir, req in enumerate(reqlistbg):
req.notify_finished(progressNotify)
req.submit()
traceLogger.debug("Requests fired")
#Fixme: Maybe later request only part of the region?
#image=self.inputs["Images"][i][:].wait()
for ir, req in enumerate(reqlistlabels):
labblock = req.wait()
image = reqlistfeat[ir].wait()
labbgblock = reqlistbg[ir].wait()
labblock = labblock.reshape((image.shape[:-1]))
image = image.reshape((-1, image.shape[-1]))
labbgindices = np.where(labbgblock == 2)
labbgindices = np.ravel_multi_index(labbgindices, labbgblock.shape)
newDot, mapping, tags = \
self._svr.prepareDataRefactored(labblock, labbgindices)
#self._svr.prepareData(labblock, smooth = True)
labels = newDot[mapping]
features = image[mapping]
featMatrix.append(features)
labelsMatrix.append(labels)
tagList.append(tags)
progress = progress_outer[0]
traceLogger.debug("Requests processed")
self.progressSignal(80 / numImages)
if len(featMatrix) == 0 or len(labelsMatrix) == 0:
result[:] = None
else:
posTags = [tag[0] for tag in tagList]
negTags = [tag[1] for tag in tagList]
numPosTags = np.sum(posTags)
numTags = np.sum(posTags) + np.sum(negTags)
fullFeatMatrix = np.ndarray((numTags, self.Images[0].meta.shape[-1]), dtype = np.float64)
fullLabelsMatrix = np.ndarray((numTags), dtype = np.float64)
fullFeatMatrix[:] = np.NAN
fullLabelsMatrix[:] = np.NAN
currPosCount = 0
currNegCount = numPosTags
for i, posCount in enumerate(posTags):
fullFeatMatrix[currPosCount:currPosCount + posTags[i],:] = featMatrix[i][:posCount,:]
fullLabelsMatrix[currPosCount:currPosCount + posTags[i]] = labelsMatrix[i][:posCount]
fullFeatMatrix[currNegCount:currNegCount + negTags[i],:] = featMatrix[i][posCount:,:]
fullLabelsMatrix[currNegCount:currNegCount + negTags[i]] = labelsMatrix[i][posCount:]
currPosCount += posTags[i]
currNegCount += negTags[i]
assert(not np.isnan(np.sum(fullFeatMatrix)))
fullTags = [np.sum(posTags), np.sum(negTags)]
#pool = RequestPool()
maxima = np.max(fullFeatMatrix, axis=0)
minima = np.min(fullFeatMatrix, axis=0)
normalizationFactors = (minima,maxima)
boxConstraintList = []
boxConstraints = None
if self.BoxConstraintRois.ready() and self.BoxConstraintValues.ready():
for i, slot in enumerate(zip(self.BoxConstraintRois,self.BoxConstraintValues)):
for constr, val in zip(slot[0].value, slot[1].value):
boxConstraintList.append((i, constr, val))
if len(boxConstraintList) > 0:
boxConstraints = self.constructBoxConstraints(boxConstraintList)
params = self._svr.get_params()
try:
pool = RequestPool()
def train_and_store(i):
result[i] = SVR(minmax = normalizationFactors, **params)
result[i].fitPrepared(fullFeatMatrix, fullLabelsMatrix, tags = fullTags, boxConstraints = boxConstraints, numRegressors
= self.numRegressors, trainAll = False)
for i in range(self.numRegressors):
req = pool.request(partial(train_and_store, i))
pool.wait()
pool.clean()
except:
logger.error("ERROR: could not learn regressor")
logger.error("fullFeatMatrix shape = {}, dtype = {}".format(fullFeatMatrix.shape, fullFeatMatrix.dtype) )
logger.error("fullLabelsMatrix shape = {}, dtype = {}".format(fullLabelsMatrix.shape, fullLabelsMatrix.dtype) )
raise
finally:
self.progressSignal(100)
return result
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 execute(self, slot, subindex, roi, result):
assert slot in [self.Predictions,
self.Probabilities,
self.CachedProbabilities,
self.ProbabilityChannels,
self.BadObjects]
times = roi._l
if len(times) == 0:
# we assume that 0-length requests are requesting everything
times = range(self.Predictions.meta.shape[0])
if slot is self.CachedProbabilities:
return {t: self.prob_cache[t] for t in times if t in self.prob_cache}
forests=self.inputs["Classifier"][:].wait()
if forests is None or forests[0] is None:
# this happens if there was no data to train with
return dict((t, numpy.array([])) for t in times)
feats = {}
prob_predictions = {}
selected = self.SelectedFeatures([]).wait()
# FIXME: self.prob_cache is shared, so we need to block.
# However, this makes prediction single-threaded.
self.lock.acquire()
try:
for t in times:
if t in self.prob_cache:
continue
tmpfeats = self.Features([t]).wait()
ftmatrix, _, col_names = make_feature_array(tmpfeats, selected)
rows, cols = replace_missing(ftmatrix)
self.bad_objects[t] = numpy.zeros((ftmatrix.shape[0],))
self.bad_objects[t][rows] = 1
feats[t] = ftmatrix
prob_predictions[t] = [0] * len(forests)
def predict_forest(_t, forest_index):
# Note: We can't use RandomForest.predictLabels() here because we're training in parallel,
# and we have to average the PROBABILITIES from all forests.
# Averaging the label predictions from each forest is NOT equivalent.
# For details please see wikipedia:
# http://en.wikipedia.org/wiki/Electoral_College_%28United_States%29#Irrelevancy_of_national_popular_vote
# (^-^)
prob_predictions[_t][forest_index] = forests[forest_index].predictProbabilities(feats[_t].astype(numpy.float32))
# predict the data with all the forests in parallel
pool = RequestPool()
for t in times:
if t in self.prob_cache:
continue
for i, f in enumerate(forests):
req = Request( partial(predict_forest, t, i) )
pool.add(req)
pool.wait()
pool.clean()
for t in times:
if t not in self.prob_cache:
# prob_predictions is a dict-of-lists-of-arrays, indexed as follows:
# prob_predictions[t][forest_index][object_index, class_index]
# Stack the forests together and average them.
stacked_predictions = numpy.array( prob_predictions[t] )
averaged_predictions = numpy.average( stacked_predictions, axis=0 )
assert averaged_predictions.shape[0] == len(feats[t])
self.prob_cache[t] = averaged_predictions
self.prob_cache[t][0] = 0 # Background probability is always zero
if slot == self.Probabilities:
return { t : self.prob_cache[t] for t in times }
elif slot == self.Predictions:
# FIXME: Support SegmentationThreshold again...
labels = dict()
for t in times:
prob_sum = numpy.sum(self.prob_cache[t], axis=1)
labels[t] = 1 + numpy.argmax(self.prob_cache[t], axis=1)
labels[t][0] = 0 # Background gets the zero label
return labels
elif slot == self.ProbabilityChannels:
try:
prob_single_channel = {t: self.prob_cache[t][:, subindex[0]]
for t in times}
except:
# no probabilities available for this class; return zeros
prob_single_channel = {t: numpy.zeros((self.prob_cache[t].shape[0], 1))
for t in times}
return prob_single_channel
elif slot == self.BadObjects:
return { t : self.bad_objects[t] for t in times }
else:
assert False, "Unknown input slot"
finally:
self.lock.release()
def execute(self, slot, subindex, roi, result):
featList = []
all_col_names = []
labelsList = []
# will be available at slot self.Warnings
all_bad_objects = defaultdict(lambda: defaultdict(list))
all_bad_feats = set()
selected = self.SelectedFeatures([]).wait()
if len(selected)==0:
# no features - no predictions
self.Classifier.setValue(None)
return
for i in range(len(self.Labels)):
# this loop is by image, not time!
# TODO: we should be able to use self.Labels[i].value,
# but the current implementation of Slot.value() does not
# do the right thing.
labels_image = self.Labels[i]([]).wait()
labels_image_filtered = {}
nztimes = []
for timestep, labels_time in labels_image.iteritems():
nz = numpy.nonzero(labels_time)
if len(nz[0])==0:
continue
else:
nztimes.append(timestep)
labels_image_filtered[timestep] = labels_time
if len(nztimes)==0:
continue
# compute the features if there are nonzero labels in this image
# and only for the time steps, which have labels
feats = self.Features[i](nztimes).wait()
featstmp, row_names, col_names, labelstmp = make_feature_array(feats, selected, labels_image_filtered)
if labelstmp.size == 0 or featstmp.size == 0:
continue
rows, cols = replace_missing(featstmp)
featList.append(featstmp)
all_col_names.append(tuple(col_names))
labelsList.append(labelstmp)
for idx in rows:
t, obj = row_names[idx]
all_bad_objects[i][t].append(obj)
for c in cols:
all_bad_feats.add(col_names[c])
if len(labelsList)==0:
#no labels, return here
self.Classifier.setValue(None)
return
self._warnBadObjects(all_bad_objects, all_bad_feats)
if not len(set(all_col_names)) == 1:
raise Exception('different time slices did not have same features.')
featMatrix = _concatenate(featList, axis=0)
labelsMatrix = _concatenate(labelsList, axis=0)
logger.info("training on matrix of shape {}".format(featMatrix.shape))
if featMatrix.size == 0 or labelsMatrix.size == 0:
result[:] = None
return
oob = [0] * self.ForestCount.value
try:
# train and store forests in parallel
pool = RequestPool()
for i in range(self.ForestCount.value):
def train_and_store(number):
result[number] = vigra.learning.RandomForest(self._tree_count)
oob[number] = result[number].learnRF(featMatrix.astype(numpy.float32), numpy.asarray(labelsMatrix, dtype=numpy.uint32))
req = Request( partial(train_and_store, i) )
pool.add( req )
pool.wait()
pool.clean()
except:
logger.warn("couldn't learn classifier")
raise
oob_total = numpy.mean(oob)
logger.info("training finished, out of bag error: {}".format(oob_total))
return result
def execute(self, slot, subindex, roi, result):
progress = 0
self.progressSignal(progress)
numImages = len(self.Images)
featMatrix=[]
labelsMatrix=[]
tagsMatrix = []
result[0] = SVR(self.UnderMult.value, self.OverMult.value, limitDensity = True, **self.SelectedOption.value)
for i,labels in enumerate(self.inputs["Labels"]):
if labels.meta.shape is not None:
#labels=labels[:].wait()
blocks = self.inputs["nonzeroLabelBlocks"][i][0].wait()
progress += 10/numImages
self.progressSignal(progress)
reqlistlabels = []
reqlistfeat = []
traceLogger.debug("Sending requests for {} non-zero blocks (labels and data)".format( len(blocks[0])) )
for b in blocks[0]:
request = labels[b]
featurekey = list(b)
featurekey[-1] = slice(None, None, None)
request2 = self.inputs["Images"][i][featurekey]
reqlistlabels.append(request)
reqlistfeat.append(request2)
traceLogger.debug("Requests prepared")
numLabelBlocks = len(reqlistlabels)
progress_outer = [progress] # Store in list for closure access
if numLabelBlocks > 0:
progressInc = (80-10)/numLabelBlocks/numImages
def progressNotify(req):
# Note: If we wanted perfect progress reporting, we could use lock here
# to protect the progress from being incremented simultaneously.
# But that would slow things down and imperfect reporting is okay for our purposes.
progress_outer[0] += progressInc/2
self.progressSignal(progress_outer[0])
for ir, req in enumerate(reqlistfeat):
image = req.notify_finished(progressNotify)
for ir, req in enumerate(reqlistlabels):
labblock = req.notify_finished(progressNotify)
traceLogger.debug("Requests fired")
for ir, req in enumerate(reqlistlabels):
traceLogger.debug("Waiting for a label block...")
labblock = req.wait()
traceLogger.debug("Waiting for an image block...")
image = reqlistfeat[ir].wait()
newImg, newDot, mapping, tags = \
result[0].prepareData(image, labblock, sigma = self.Sigma.value, smooth = True, normalize = False)
features = newImg[mapping]
labbla = newDot[mapping]
#indexes=np.nonzero(labblock[...,0].view(np.ndarray))
#features=image[indexes]
#labbla=labblock[indexes]
featMatrix.append(features)
labelsMatrix.append(labbla)
tagsMatrix.append(tags)
progress = progress_outer[0]
traceLogger.debug("Requests processed")
self.progressSignal(80/numImages)
if len(featMatrix) == 0 or len(labelsMatrix) == 0:
# If there was no actual data for the random forest to train with, we return None
result[:] = None
else:
featMatrix=np.concatenate(featMatrix,axis=0)
labelsMatrix=np.concatenate(labelsMatrix,axis=0)
tagsMatrix=np.concatenate(tagsMatrix,axis=0)
try:
logger.debug("Learning with Vigra...")
pool = RequestPool()
#result[0].fitPrepared(featMatrix, labelsMatrix, tagsMatrix, self.Epsilon.value)
req = pool.request(partial(result[0].fitPrepared, featMatrix, labelsMatrix, tagsMatrix, self.Epsilon.value))
pool.wait()
pool.clean()
logger.debug("Vigra finished")
except:
logger.error( "ERROR: could not learn classifier" )
logger.error( "featMatrix shape={}, max={}, dtype={}".format(featMatrix.shape, featMatrix.max(), featMatrix.dtype) )
logger.error( "labelsMatrix shape={}, max={}, dtype={}".format(labelsMatrix.shape, labelsMatrix.max(), labelsMatrix.dtype ) )
raise
finally:
self.progressSignal(100)
return result
def extractHistograms(volume, labels, patchSize=64, haloSize=0,
nBins=30, intRange=(0, 255), appendPositions=False):
'''
extracts histograms from 3d-volume
- labels are
0 ignore
1 positive
2 negative
- histogram extraction is attempted to be done in parallel
- patches that intersect with the volume border are discarded
- volume and labels must be 3d, and in order 'zyx' (if not VigraArrays)
- returns: np.ndarray, shape: (nSamples,nBins+1), last column is the label
'''
# progress reporter class, histogram extraction can take quite a long time
class ProgressReporter(object):
lock = None
def __init__(self, nThreads):
self.lock = ThreadLock()
self.nThreads = nThreads
self.status = np.zeros((nThreads,))
def report(self, index):
self.lock.acquire()
self.status[index] = 1
logger.debug("Finished threads: %d/%d." %
(self.status.sum(), len(self.status)))
self.lock.release()
# sanity checks
assert len(volume.shape) == 3, "Volume must be 3d data"
assert volume.shape == labels.shape,\
"Volume and labels must have the same shape"
try:
volumeZYX = volume.withAxes(*'zyx')
labelsZYX = labels.withAxes(*'zyx')
except AttributeError:
# can't blame me
volumeZYX = volume
labelsZYX = labels
pass
# compute actual patch size
patchSize = patchSize + 2*haloSize
# fill list of patch centers (VigraArray does not support bitwise_or)
ind_z, ind_y, ind_x = np.where(
(labelsZYX == 1).view(np.ndarray) | (labelsZYX == 2).view(np.ndarray))
index = np.arange(len(ind_z))
# prepare chunking of histogram centers
chunkSize = 10000 # FIXME magic number??
nChunks = len(index)//chunkSize + (1 if len(index) % chunkSize > 0 else 0)
sliceList = [slice(k*chunkSize, min((k+1)*chunkSize, len(index)))
for k in range(nChunks)]
histoList = [None]*nChunks
# prepare subroutine for parallel extraction
reporter = ProgressReporter(nChunks)
#BEGIN subroutine
def _extractHistogramsSub(itemList):
xs = ind_x[itemList]
ys = ind_y[itemList]
zs = ind_z[itemList]
ymin = ys - patchSize//2
ymax = ymin + patchSize
xmin = xs - patchSize//2
xmax = xmin + patchSize
validPatchIndices = np.where(
np.all(
(ymin >= 0,
xmin >= 0,
xmax <= volumeZYX.shape[2],
ymax <= volumeZYX.shape[1]),
axis=0))[0]
if appendPositions:
out = np.zeros((len(validPatchIndices), nBins+4))
else:
out = np.zeros((len(validPatchIndices), nBins+1))
for k, patchInd in enumerate(validPatchIndices):
x = xs[patchInd]
y = ys[patchInd]
z = zs[patchInd]
vol = volumeZYX[z, ymin[patchInd]:ymax[patchInd],
xmin[patchInd]:xmax[patchInd]]
(out[k, :nBins], _) = np.histogram(
vol, bins=nBins, range=intRange, density=True)
out[k, nBins] = 1 if labelsZYX[z, y, x] == 1 else 0
if appendPositions:
out[k, nBins+1:] = [z, y, x]
return out
def partFun(i):
itemList = index[sliceList[i]]
histos = _extractHistogramsSub(itemList)
histoList[i] = histos
reporter.report(i)
#END subroutine
# pool the extraction requests
pool = RequestPool()
for i in range(nChunks):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
return np.vstack(histoList)
def _felzenszwalbTraining(self, negative, positive):
'''
we want to train on a 'hard' subset of the training data, see
FELZENSZWALB ET AL.: OBJECT DETECTION WITH DISCRIMINATIVELY TRAINED PART-BASED MODELS (4.4), PAMI 32/9
'''
#TODO sanity checks
n = (self.PatchSize.value + self.HaloSize.value)**2
method = self.DetectionMethod.value
# set options for Felzenszwalb training
firstSamples = self._felzenOpts["firstSamples"]
maxRemovePerStep = self._felzenOpts["maxRemovePerStep"]
maxAddPerStep = self._felzenOpts["maxAddPerStep"]
maxSamples = self._felzenOpts["maxSamples"]
nTrainingSteps = self._felzenOpts["nTrainingSteps"]
# initial choice of training samples
(initNegative, choiceNegative, _, _) = \
_chooseRandomSubset(negative, min(firstSamples, len(negative)))
(initPositive, choicePositive, _, _) = \
_chooseRandomSubset(positive, min(firstSamples, len(positive)))
# setup for parallel training
samples = [negative, positive]
choice = [choiceNegative, choicePositive]
S_t = [initNegative, initPositive]
finished = [False, False]
### BEGIN SUBROUTINE ###
def felzenstep(x, cache, ind):
case = ("positive" if ind > 0 else "negative") + " set"
pred = self.predict(x, method=method)
hard = np.where(pred != ind)[0]
easy = np.setdiff1d(range(len(x)), hard)
logger.debug(" {}: currently {} hard and {} easy samples".format(
case, len(hard), len(easy)))
# shrink the cache
easyInCache = np.intersect1d(easy, cache) if len(easy) > 0 else []
if len(easyInCache) > 0:
(removeFromCache, _, _, _) = _chooseRandomSubset(
easyInCache, min(len(easyInCache), maxRemovePerStep))
cache = np.setdiff1d(cache, removeFromCache)
logger.debug(" {}: shrunk the cache by {} elements".format(
case, len(removeFromCache)))
# grow the cache
temp = len(cache)
addToCache = _chooseRandomSubset(
hard, min(len(hard), maxAddPerStep))[0]
cache = np.union1d(cache, addToCache)
addedHard = len(cache)-temp
logger.debug(" {}: grown the cache by {} elements".format(
case, addedHard))
if len(cache) > maxSamples:
logger.debug(
" {}: Cache to big, removing elements.".format(case))
cache = _chooseRandomSubset(cache, maxSamples)[0]
# apply the cache
C = x[cache]
return (C, cache, addedHard == 0)
### END SUBROUTINE ###
### BEGIN PARALLELIZATION FUNCTION ###
def partFun(i):
(C, newChoice, newFinished) = felzenstep(samples[i], choice[i], i)
S_t[i] = C
choice[i] = newChoice
finished[i] = newFinished
### END PARALLELIZATION FUNCTION ###
for k in range(nTrainingSteps):
logger.debug(
"Felzenszwalb Training " +
"(step {}/{}): {} hard negative samples, {}".format(
k+1, nTrainingSteps, len(S_t[0]), len(S_t[1])) +
"hard positive samples.")
self.fit(S_t[0], S_t[1], method=method)
pool = RequestPool()
for i in range(len(S_t)):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
if np.all(finished):
#already have all hard examples in training set
break
self.fit(S_t[0], S_t[1], method=method)
logger.debug(" Finished Felzenszwalb Training.")
def execute(self, slot, subindex, roi, result):
assert slot in [
self.Predictions, self.Probabilities, self.CachedProbabilities,
self.ProbabilityChannels, self.BadObjects
]
times = roi._l
if len(times) == 0:
# we assume that 0-length requests are requesting everything
times = range(self.Predictions.meta.shape[0])
if slot is self.CachedProbabilities:
return {
t: self.prob_cache[t]
for t in times if t in self.prob_cache
}
forests = self.inputs["Classifier"][:].wait()
if forests is None or forests[0] is None:
# this happens if there was no data to train with
return dict((t, numpy.array([])) for t in times)
feats = {}
prob_predictions = {}
selected = self.SelectedFeatures([]).wait()
# FIXME: self.prob_cache is shared, so we need to block.
# However, this makes prediction single-threaded.
self.lock.acquire()
try:
for t in times:
if t in self.prob_cache:
continue
tmpfeats = self.Features([t]).wait()
ftmatrix, _, col_names = make_feature_array(tmpfeats, selected)
rows, cols = replace_missing(ftmatrix)
self.bad_objects[t] = numpy.zeros((ftmatrix.shape[0], ))
self.bad_objects[t][rows] = 1
feats[t] = ftmatrix
prob_predictions[t] = [0] * len(forests)
def predict_forest(_t, forest_index):
# Note: We can't use RandomForest.predictLabels() here because we're training in parallel,
# and we have to average the PROBABILITIES from all forests.
# Averaging the label predictions from each forest is NOT equivalent.
# For details please see wikipedia:
# http://en.wikipedia.org/wiki/Electoral_College_%28United_States%29#Irrelevancy_of_national_popular_vote
# (^-^)
prob_predictions[_t][forest_index] = forests[
forest_index].predictProbabilities(feats[_t].astype(
numpy.float32))
# predict the data with all the forests in parallel
pool = RequestPool()
for t in times:
if t in self.prob_cache:
continue
for i, f in enumerate(forests):
req = Request(partial(predict_forest, t, i))
pool.add(req)
pool.wait()
pool.clean()
for t in times:
if t not in self.prob_cache:
# prob_predictions is a dict-of-lists-of-arrays, indexed as follows:
# prob_predictions[t][forest_index][object_index, class_index]
# Stack the forests together and average them.
stacked_predictions = numpy.array(prob_predictions[t])
averaged_predictions = numpy.average(stacked_predictions,
axis=0)
assert averaged_predictions.shape[0] == len(feats[t])
self.prob_cache[t] = averaged_predictions
self.prob_cache[t][
0] = 0 # Background probability is always zero
if slot == self.Probabilities:
return {t: self.prob_cache[t] for t in times}
elif slot == self.Predictions:
# FIXME: Support SegmentationThreshold again...
labels = dict()
for t in times:
prob_sum = numpy.sum(self.prob_cache[t], axis=1)
labels[t] = 1 + numpy.argmax(self.prob_cache[t], axis=1)
labels[t][0] = 0 # Background gets the zero label
return labels
elif slot == self.ProbabilityChannels:
try:
prob_single_channel = {
t: self.prob_cache[t][:, subindex[0]]
for t in times
}
except:
# no probabilities available for this class; return zeros
prob_single_channel = {
t: numpy.zeros((self.prob_cache[t].shape[0], 1))
for t in times
}
return prob_single_channel
elif slot == self.BadObjects:
return {t: self.bad_objects[t] for t in times}
else:
assert False, "Unknown input slot"
finally:
self.lock.release()
def execute(self, slot, subindex, slot_roi, target):
assert slot == self.Features or slot == self.Output
if slot == self.Features:
feature_slice = roiToSlice(slot_roi.start, slot_roi.stop)
index = subindex[0]
feature_slice = list(feature_slice)
# Translate channel slice of this feature to the channel slice of the output slot.
output_channel_offset = self.featureOutputChannels[index][0]
feature_slice[1] = slice(
output_channel_offset + feature_slice[1].start,
output_channel_offset + feature_slice[1].stop)
slot_roi = SubRegion(self.Output, pslice=feature_slice)
# Get output slot region for this channel
return self.execute(self.Output, (), slot_roi, target)
elif slot == self.Output:
# Correlation of variable 'families' representing reference frames:
# ______________________________
# | input/output frame | input/output shape given by slots
# | _________________________ |
# | | smooth frame | | pre-smoothing op needs halo around filter roi
# | | ____________________ | |
# | | |filter frame | | | filter needs halo around target roi
# | | | _______________ | | |
# | | | | target frame | | | | target is given by output_roi
# note: The 'full_' variable prefix refers to the full 5D shape (tczyx), without 'full_' variables mostly
# refer to the 3D space subregion (zyx)
full_output_slice = slot_roi.toSlice()
logger.debug(
f"OpPixelFeaturesPresmoothed: request {slot_roi.pprint()}")
assert (slot_roi.stop <= self.Output.meta.shape).all()
full_output_shape = self.Output.meta.shape
full_output_start, full_output_stop = sliceToRoi(
full_output_slice, full_output_shape)
assert len(full_output_shape) == 5
if all(self.ComputeIn2d.value
): # todo: check for this particular slice
axes2enlarge = (0, 1, 1)
else:
axes2enlarge = (1, 1, 1)
output_shape = full_output_shape[2:]
output_start = full_output_start[2:]
output_stop = full_output_stop[2:]
axistags = self.Output.meta.axistags
target = target.view(vigra.VigraArray)
target.axistags = copy.copy(axistags)
# filter roi in input frame
# sigma = 0.7, because the features receive a pre-smoothed array and don't need much of a neighborhood
input_filter_start, input_filter_stop = roi.enlargeRoiForHalo(
output_start,
output_stop,
output_shape,
0.7,
self.WINDOW_SIZE,
enlarge_axes=axes2enlarge)
# smooth roi in input frame
input_smooth_start, input_smooth_stop = roi.enlargeRoiForHalo(
input_filter_start,
input_filter_stop,
output_shape,
self.max_sigma,
self.WINDOW_SIZE,
enlarge_axes=axes2enlarge,
)
# target roi in filter frame
filter_target_start = roi.TinyVector(output_start -
input_filter_start)
filter_target_stop = roi.TinyVector(output_stop -
input_filter_start)
# filter roi in smooth frame
smooth_filter_start = roi.TinyVector(input_filter_start -
input_smooth_start)
smooth_filter_stop = roi.TinyVector(input_filter_stop -
input_smooth_start)
filter_target_slice = roi.roiToSlice(filter_target_start,
filter_target_stop)
input_smooth_slice = roi.roiToSlice(input_smooth_start,
input_smooth_stop)
# pre-smooth for all requested time slices and all channels
full_input_smooth_slice = (full_output_slice[0], slice(None),
*input_smooth_slice)
req = self.Input[full_input_smooth_slice]
source = req.wait()
req.clean()
req.destination = None
if source.dtype != numpy.float32:
sourceF = source.astype(numpy.float32)
try:
source.resize((1, ), refcheck=False)
except Exception:
pass
del source
source = sourceF
sourceV = source.view(vigra.VigraArray)
sourceV.axistags = copy.copy(self.Input.meta.axistags)
dimCol = len(self.scales)
dimRow = self.matrix.shape[0]
presmoothed_source = [None] * dimCol
source_smooth_shape = tuple(smooth_filter_stop -
smooth_filter_start)
full_source_smooth_shape = (
full_output_stop[0] - full_output_start[0],
self.Input.meta.shape[1],
) + source_smooth_shape
try:
for j in range(dimCol):
for i in range(dimRow):
if self.matrix[i, j]:
# There is at least one filter op with this scale
break
else:
# There is no filter op at this scale
continue
if self.scales[j] > 1.0:
tempSigma = math.sqrt(self.scales[j]**2 - 1.0)
else:
tempSigma = self.scales[j]
presmoothed_source[j] = numpy.ndarray(
full_source_smooth_shape, numpy.float32)
droi = (
(0, *tuple(smooth_filter_start._asint())),
(sourceV.shape[1],
*tuple(smooth_filter_stop._asint())),
)
for i, vsa in enumerate(sourceV.timeIter()):
presmoothed_source[j][
i, ...] = self._computeGaussianSmoothing(
vsa,
tempSigma,
droi,
in2d=self.ComputeIn2d.value[j])
except RuntimeError as e:
if "kernel longer than line" in str(e):
raise RuntimeError(
"Feature computation error:\nYour image is too small to apply a filter with "
f"sigma={self.scales[j]:.1f}. Please select features with smaller sigmas."
)
else:
raise e
del sourceV
try:
source.resize((1, ), refcheck=False)
except ValueError:
# Sometimes this fails, but that's okay.
logger.debug("Failed to free array memory.")
del source
cnt = 0
written = 0
closures = []
# connect individual operators
for i in range(dimRow):
for j in range(dimCol):
if self.matrix[i, j]:
oslot = self.featureOps[i][j].Output
req = None
slices = oslot.meta.shape[1]
if (cnt + slices >= slot_roi.start[1]
and slot_roi.start[1] - cnt < slices
and slot_roi.start[1] + written <
slot_roi.stop[1]):
begin = 0
if cnt < slot_roi.start[1]:
begin = slot_roi.start[1] - cnt
end = slices
if cnt + end > slot_roi.stop[1]:
end = slot_roi.stop[1] - cnt
# feature slice in output frame
feature_slice = (slice(None),
slice(
written, written + end -
begin)) + (slice(None), ) * 3
subtarget = target[feature_slice]
# readjust the roi for the new source array
full_filter_target_slice = [
full_output_slice[0],
slice(begin, end), *filter_target_slice
]
filter_target_roi = SubRegion(
oslot, pslice=full_filter_target_slice)
closure = partial(
oslot.operator.execute,
oslot,
(),
filter_target_roi,
subtarget,
sourceArray=presmoothed_source[j],
)
closures.append(closure)
written += end - begin
cnt += slices
pool = RequestPool()
for c in closures:
pool.request(c)
pool.wait()
pool.clean()
for i in range(len(presmoothed_source)):
if presmoothed_source[i] is not None:
try:
presmoothed_source[i].resize((1, ))
except Exception:
presmoothed_source[i] = None
def execute(self, slot, subindex, roi, result):
featMatrix = []
labelsMatrix = []
for i in range(len(self.Labels)):
feats = self.Features[i]([]).wait()
# TODO: we should be able to use self.Labels[i].value,
# but the current implementation of Slot.value() does not
# do the right thing.
labels = self.Labels[i]([]).wait()
for t in sorted(feats.keys()):
featsMatrix_tmp = []
labelsMatrix_tmp = []
lab = labels[t].squeeze()
index = numpy.nonzero(lab)
labelsMatrix_tmp.append(lab[index])
#check that all requested features are present
for featname in config.selected_features:
for channel in feats[t]:
if not featname in channel.keys():
print "Feature", featname, "has not been computed in the previous step"
print "We only have the following features now:", channel.keys()
result[:] = None
return
else:
value = channel[featname]
ft = numpy.asarray(value.squeeze())
featsMatrix_tmp.append(ft[index])
featMatrix.append(_concatenate(featsMatrix_tmp, axis=1))
labelsMatrix.append(_concatenate(labelsMatrix_tmp, axis=1))
featMatrix = _concatenate(featMatrix, axis=0)
labelsMatrix = _concatenate(labelsMatrix, axis=0)
print "training on matrix:", featMatrix.shape
if len(featMatrix) == 0 or len(labelsMatrix) == 0:
result[:] = None
return
oob = [0]*self.ForestCount.value
try:
# Ensure there are no NaNs in the feature matrix
# TODO: There should probably be a better way to fix this...
featMatrix = numpy.asarray(featMatrix, dtype=numpy.float32)
nanFeatMatrix = numpy.isnan(featMatrix)
if nanFeatMatrix.any():
warnings.warn("Feature matrix has NaN values! Replacing with 0.0...")
featMatrix[numpy.where(nanFeatMatrix)] = 0.0
# train and store forests in parallel
pool = RequestPool()
for i in range(self.ForestCount.value):
def train_and_store(number):
result[number] = vigra.learning.RandomForest(self._tree_count)
oob[number] = result[number].learnRF(featMatrix, numpy.asarray(labelsMatrix, dtype=numpy.uint32))
print "intermediate oob:", oob[number]
req = Request( partial(train_and_store, i) )
pool.add( req )
pool.wait()
pool.clean()
except:
print ("couldn't learn classifier")
raise
oob_total = numpy.mean(oob)
print "training finished, out of bag error:", oob_total
return result
def _executeOutput(self, slot, subindex, roi, result):
t = time.time()
key = roi.toSlice()
shape = self.Output.meta.shape
start, stop = sliceToRoi(key, shape)
self._lock.acquire()
ch = self._cacheHits
ch += 1
self._cacheHits = ch
self._running += 1
if self._cache is None:
self._allocateCache()
cacheView = self._cache[:] #prevent freeing of cache during running this function
blockStart = (1.0 * start / self._blockShape).floor()
blockStop = (1.0 * stop / self._blockShape).ceil()
blockKey = roiToSlice(blockStart, blockStop)
blockSet = self._blockState[blockKey]
# this is a little optimization to shortcut
# many lines of python code when all data is
# is already in the cache:
if numpy.logical_or(blockSet == OpArrayCache.CLEAN,
blockSet == OpArrayCache.FIXED_DIRTY).all():
result[:] = self._cache[roiToSlice(start, stop)]
self._running -= 1
self._updatePriority()
cacheView = None
self._lock.release()
return
inProcessQueries = numpy.unique(
numpy.extract(blockSet == OpArrayCache.IN_PROCESS,
self._blockQuery[blockKey]))
cond = (blockSet == OpArrayCache.DIRTY)
tileWeights = fastWhere(cond, 1, 128**3, numpy.uint32)
trueDirtyIndices = numpy.nonzero(cond)
tileArray = drtile.test_DRTILE(tileWeights, 128**3).swapaxes(0, 1)
dirtyRois = []
half = tileArray.shape[0] / 2
dirtyPool = RequestPool()
for i in range(tileArray.shape[1]):
drStart3 = tileArray[:half, i]
drStop3 = tileArray[half:, i]
drStart2 = drStart3 + blockStart
drStop2 = drStop3 + blockStart
drStart = drStart2 * self._blockShape
drStop = drStop2 * self._blockShape
shape = self.Output.meta.shape
drStop = numpy.minimum(drStop, shape)
drStart = numpy.minimum(drStart, shape)
key3 = roiToSlice(drStart3, drStop3)
key2 = roiToSlice(drStart2, drStop2)
key = roiToSlice(drStart, drStop)
if not self._fixed:
dirtyRois.append([drStart, drStop])
req = self.inputs["Input"][key].writeInto(self._cache[key])
req.uncancellable = True #FIXME
dirtyPool.add(req)
self._blockQuery[key2] = weakref.ref(req)
#sanity check:
if (self._blockState[key2] != OpArrayCache.DIRTY).any():
logger.warning("original condition" + str(cond))
logger.warning("original tilearray {} {}".format(
tileArray, tileArray.shape))
logger.warning("original tileWeights {} {}".format(
tileWeights, tileWeights.shape))
logger.warning("sub condition {}".format(
self._blockState[key2] == OpArrayCache.DIRTY))
logger.warning("START={}, STOP={}".format(
drStart2, drStop2))
import h5py
with h5py.File("test.h5", "w") as f:
f.create_dataset("data", data=tileWeights)
logger.warning(
"%r \n %r \n %r\n %r\n %r \n%r" %
(key2, blockKey, self._blockState[key2],
self._blockState[blockKey][trueDirtyIndices],
self._blockState[blockKey], tileWeights))
assert False
self._blockState[key2] = OpArrayCache.IN_PROCESS
# indicate the inprocessing state, by setting array to 0 (i.e. IN_PROCESS)
if not self._fixed:
blockSet[:] = fastWhere(cond, OpArrayCache.IN_PROCESS, blockSet,
numpy.uint8)
else:
# Someone asked for some dirty blocks while we were fixed.
# Mark these blocks to be signaled as dirty when we become unfixed
blockSet[:] = fastWhere(cond, OpArrayCache.FIXED_DIRTY, blockSet,
numpy.uint8)
self._has_fixed_dirty_blocks = True
self._lock.release()
temp = itertools.count(0)
#wait for all requests to finish
dirtyPool.wait()
if len(dirtyPool) > 0:
# Signal that something was updated.
# Note that we don't need to do this for the 'in process' queries (below)
# because they are already in the dirtyPool in some other thread
self.Output._sig_value_changed()
dirtyPool.clean()
# indicate the finished inprocess state (i.e. CLEAN)
if not self._fixed and temp.next() == 0:
with self._lock:
blockSet[:] = fastWhere(cond, OpArrayCache.CLEAN, blockSet,
numpy.uint8)
self._blockQuery[blockKey] = fastWhere(
cond, None, self._blockQuery[blockKey], object)
inProcessPool = RequestPool()
#wait for all in process queries
for req in inProcessQueries:
req = req() # get original req object from weakref
if req is not None:
inProcessPool.add(req)
inProcessPool.wait()
inProcessPool.clean()
# finally, store results in result area
self._lock.acquire()
if self._cache is not None:
result[:] = self._cache[roiToSlice(start, stop)]
else:
self.inputs["Input"][roiToSlice(start,
stop)].writeInto(result).wait()
self._running -= 1
self._updatePriority()
cacheView = None
self._lock.release()
self.logger.debug("read %s took %f sec." %
(roi.pprint(), time.time() - t))
def execute(self, slot, subindex, roi, result):
featList = []
all_col_names = []
labelsList = []
# will be available at slot self.Warnings
all_bad_objects = defaultdict(lambda: defaultdict(list))
all_bad_feats = set()
selected = self.SelectedFeatures([]).wait()
if len(selected) == 0:
# no features - no predictions
self.Classifier.setValue(None)
return
for i in range(len(self.Labels)):
# FIXME: we should only compute the features if there are nonzero labels in this image
feats = self.Features[i]([]).wait()
# TODO: we should be able to use self.Labels[i].value,
# but the current implementation of Slot.value() does not
# do the right thing.
labels = self.Labels[i]([]).wait()
featstmp, row_names, col_names, labelstmp = make_feature_array(
feats, selected, labels)
if labelstmp.size == 0 or featstmp.size == 0:
continue
rows, cols = replace_missing(featstmp)
featList.append(featstmp)
all_col_names.append(tuple(col_names))
labelsList.append(labelstmp)
for idx in rows:
t, obj = row_names[idx]
all_bad_objects[i][t].append(obj)
for c in cols:
all_bad_feats.add(col_names[c])
if len(labelsList) == 0:
#no labels, return here
self.Classifier.setValue(None)
return
self._warnBadObjects(all_bad_objects, all_bad_feats)
if not len(set(all_col_names)) == 1:
raise Exception(
'different time slices did not have same features.')
featMatrix = _concatenate(featList, axis=0)
labelsMatrix = _concatenate(labelsList, axis=0)
logger.info("training on matrix of shape {}".format(featMatrix.shape))
if featMatrix.size == 0 or labelsMatrix.size == 0:
result[:] = None
return
oob = [0] * self.ForestCount.value
try:
# train and store forests in parallel
pool = RequestPool()
for i in range(self.ForestCount.value):
def train_and_store(number):
result[number] = vigra.learning.RandomForest(
self._tree_count)
oob[number] = result[number].learnRF(
featMatrix.astype(numpy.float32),
numpy.asarray(labelsMatrix, dtype=numpy.uint32))
req = Request(partial(train_and_store, i))
pool.add(req)
pool.wait()
pool.clean()
except:
logger.warn("couldn't learn classifier")
raise
oob_total = numpy.mean(oob)
logger.info(
"training finished, out of bag error: {}".format(oob_total))
return result
def _executeOutput(self, slot, subindex, roi, result):
t = time.time()
key = roi.toSlice()
shape = self.Output.meta.shape
start, stop = sliceToRoi(key, shape)
with self._lock:
ch = self._cacheHits
ch += 1
self._cacheHits = ch
self._running += 1
if self._cache is None:
self._allocateCache()
cacheView = self._cache[:] #prevent freeing of cache during running this function
blockStart = (1.0 * start / self._blockShape).floor()
blockStop = (1.0 * stop / self._blockShape).ceil()
blockKey = roiToSlice(blockStart,blockStop)
blockSet = self._blockState[blockKey]
# this is a little optimization to shortcut
# many lines of python code when all data is
# is already in the cache:
if numpy.logical_or(blockSet == OpArrayCache.CLEAN, blockSet == OpArrayCache.FIXED_DIRTY).all():
result[:] = self._cache[roiToSlice(start, stop)]
self._running -= 1
self._updatePriority()
cacheView = None
return
inProcessQueries = numpy.unique(numpy.extract( blockSet == OpArrayCache.IN_PROCESS, self._blockQuery[blockKey]))
cond = (blockSet == OpArrayCache.DIRTY)
tileWeights = fastWhere(cond, 1, 128**3, numpy.uint32)
trueDirtyIndices = numpy.nonzero(cond)
tileArray = drtile.test_DRTILE(tileWeights, 128**3).swapaxes(0,1)
dirtyRois = []
half = tileArray.shape[0]/2
dirtyPool = RequestPool()
for i in range(tileArray.shape[1]):
drStart3 = tileArray[:half,i]
drStop3 = tileArray[half:,i]
drStart2 = drStart3 + blockStart
drStop2 = drStop3 + blockStart
drStart = drStart2*self._blockShape
drStop = drStop2*self._blockShape
shape = self.Output.meta.shape
drStop = numpy.minimum(drStop, shape)
drStart = numpy.minimum(drStart, shape)
key3 = roiToSlice(drStart3,drStop3)
key2 = roiToSlice(drStart2,drStop2)
key = roiToSlice(drStart,drStop)
if not self._fixed:
dirtyRois.append([drStart,drStop])
req = self.inputs["Input"][key].writeInto(self._cache[key])
req.uncancellable = True #FIXME
dirtyPool.add(req)
self._blockQuery[key2] = weakref.ref(req)
#sanity check:
if (self._blockState[key2] != OpArrayCache.DIRTY).any():
logger.warning( "original condition" + str(cond) )
logger.warning( "original tilearray {} {}".format( tileArray, tileArray.shape ) )
logger.warning( "original tileWeights {} {}".format( tileWeights, tileWeights.shape ) )
logger.warning( "sub condition {}".format( self._blockState[key2] == OpArrayCache.DIRTY ) )
logger.warning( "START={}, STOP={}".format( drStart2, drStop2 ) )
import h5py
with h5py.File("test.h5", "w") as f:
f.create_dataset("data",data = tileWeights)
logger.warning( "%r \n %r \n %r\n %r\n %r \n%r" % (key2, blockKey,self._blockState[key2], self._blockState[blockKey][trueDirtyIndices],self._blockState[blockKey],tileWeights) )
assert False
self._blockState[key2] = OpArrayCache.IN_PROCESS
# indicate the inprocessing state, by setting array to 0 (i.e. IN_PROCESS)
if not self._fixed:
blockSet[:] = fastWhere(cond, OpArrayCache.IN_PROCESS, blockSet, numpy.uint8)
else:
# Someone asked for some dirty blocks while we were fixed.
# Mark these blocks to be signaled as dirty when we become unfixed
blockSet[:] = fastWhere(cond, OpArrayCache.FIXED_DIRTY, blockSet, numpy.uint8)
self._has_fixed_dirty_blocks = True
temp = itertools.count(0)
#wait for all requests to finish
dirtyPool.wait()
if len( dirtyPool ) > 0:
# Signal that something was updated.
# Note that we don't need to do this for the 'in process' queries (below)
# because they are already in the dirtyPool in some other thread
self.Output._sig_value_changed()
dirtyPool.clean()
# indicate the finished inprocess state (i.e. CLEAN)
if not self._fixed and temp.next() == 0:
with self._lock:
blockSet[:] = fastWhere(cond, OpArrayCache.CLEAN, blockSet, numpy.uint8)
self._blockQuery[blockKey] = fastWhere(cond, None, self._blockQuery[blockKey], object)
inProcessPool = RequestPool()
#wait for all in process queries
for req in inProcessQueries:
req = req() # get original req object from weakref
if req is not None:
inProcessPool.add(req)
inProcessPool.wait()
inProcessPool.clean()
# finally, store results in result area
with self._lock:
if self._cache is not None:
result[:] = self._cache[roiToSlice(start, stop)]
else:
self.inputs["Input"][roiToSlice(start, stop)].writeInto(result).wait()
self._running -= 1
self._updatePriority()
cacheView = None
self.logger.debug("read %s took %f sec." % (roi.pprint(), time.time()-t))
def execute(self, slot, subindex, roi, result):
progress = 0
self.progressSignal(progress)
numImages = len(self.Images)
key = roi.toSlice()
featMatrix=[]
labelsMatrix=[]
for i,labels in enumerate(self.inputs["Labels"]):
if labels.meta.shape is not None:
#labels=labels[:].wait()
blocks = self.inputs["nonzeroLabelBlocks"][i][0].wait()
progress += 10/numImages
self.progressSignal(progress)
reqlistlabels = []
reqlistfeat = []
traceLogger.debug("Sending requests for {} non-zero blocks (labels and data)".format( len(blocks[0])) )
for b in blocks[0]:
request = labels[b]
featurekey = list(b)
featurekey[-1] = slice(None, None, None)
request2 = self.inputs["Images"][i][featurekey]
reqlistlabels.append(request)
reqlistfeat.append(request2)
traceLogger.debug("Requests prepared")
numLabelBlocks = len(reqlistlabels)
progress_outer = [progress] # Store in list for closure access
if numLabelBlocks > 0:
progressInc = (80-10)/numLabelBlocks/numImages
def progressNotify(req):
# Note: If we wanted perfect progress reporting, we could use lock here
# to protect the progress from being incremented simultaneously.
# But that would slow things down and imperfect reporting is okay for our purposes.
progress_outer[0] += progressInc/2
self.progressSignal(progress_outer[0])
for ir, req in enumerate(reqlistfeat):
image = req.notify_finished(progressNotify)
for ir, req in enumerate(reqlistlabels):
labblock = req.notify_finished(progressNotify)
traceLogger.debug("Requests fired")
for ir, req in enumerate(reqlistlabels):
traceLogger.debug("Waiting for a label block...")
labblock = req.wait()
traceLogger.debug("Waiting for an image block...")
image = reqlistfeat[ir].wait()
indexes=numpy.nonzero(labblock[...,0].view(numpy.ndarray))
features=image[indexes]
labbla=labblock[indexes]
featMatrix.append(features)
labelsMatrix.append(labbla)
progress = progress_outer[0]
traceLogger.debug("Requests processed")
self.progressSignal(80/numImages)
if len(featMatrix) == 0 or len(labelsMatrix) == 0:
# If there was no actual data for the random forest to train with, we return None
result[:] = None
else:
featMatrix=numpy.concatenate(featMatrix,axis=0)
labelsMatrix=numpy.concatenate(labelsMatrix,axis=0)
maxLabel = self.inputs["MaxLabel"].value
labelList = range(1, maxLabel+1) if maxLabel > 0 else list()
try:
logger.debug("Learning with Vigra...")
# train and store self._forest_count forests in parallel
pool = RequestPool()
for i in range(self._forest_count):
def train_and_store(number):
result[number] = vigra.learning.RandomForest(self._tree_count, labels=labelList)
result[number].learnRF( numpy.asarray(featMatrix, dtype=numpy.float32),
numpy.asarray(labelsMatrix, dtype=numpy.uint32))
req = pool.request(partial(train_and_store, i))
pool.wait()
pool.clean()
logger.debug("Vigra finished")
except:
logger.error( "ERROR: could not learn classifier" )
logger.error( "featMatrix shape={}, max={}, dtype={}".format(featMatrix.shape, featMatrix.max(), featMatrix.dtype) )
logger.error( "labelsMatrix shape={}, max={}, dtype={}".format(labelsMatrix.shape, labelsMatrix.max(), labelsMatrix.dtype ) )
raise
finally:
self.progressSignal(100)
return result
def execute(self, slot, subindex, slot_roi, target):
assert slot == self.Features or slot == self.Output
if slot == self.Features:
feature_slice = roiToSlice(slot_roi.start, slot_roi.stop)
index = subindex[0]
feature_slice = list(feature_slice)
# Translate channel slice of this feature to the channel slice of the output slot.
output_channel_offset = self.featureOutputChannels[index][0]
feature_slice[1] = slice(
output_channel_offset + feature_slice[1].start, output_channel_offset + feature_slice[1].stop
)
slot_roi = SubRegion(self.Output, pslice=feature_slice)
# Get output slot region for this channel
return self.execute(self.Output, (), slot_roi, target)
elif slot == self.Output:
# Correlation of variable 'families' representing reference frames:
# ______________________________
# | input/output frame | input/output shape given by slots
# | _________________________ |
# | | smooth frame | | pre-smoothing op needs halo around filter roi
# | | ____________________ | |
# | | |filter frame | | | filter needs halo around target roi
# | | | _______________ | | |
# | | | | target frame | | | | target is given by output_roi
# note: The 'full_' variable prefix refers to the full 5D shape (tczyx), without 'full_' variables mostly
# refer to the 3D space subregion (zyx)
full_output_slice = slot_roi.toSlice()
logger.debug(f"OpPixelFeaturesPresmoothed: request {slot_roi.pprint()}")
assert (slot_roi.stop <= self.Output.meta.shape).all()
full_output_shape = self.Output.meta.shape
full_output_start, full_output_stop = sliceToRoi(full_output_slice, full_output_shape)
assert len(full_output_shape) == 5
if all(self.ComputeIn2d.value): # todo: check for this particular slice
axes2enlarge = (0, 1, 1)
else:
axes2enlarge = (1, 1, 1)
output_shape = full_output_shape[2:]
output_start = full_output_start[2:]
output_stop = full_output_stop[2:]
axistags = self.Output.meta.axistags
target = target.view(vigra.VigraArray)
target.axistags = copy.copy(axistags)
# filter roi in input frame
# sigma = 0.7, because the features receive a pre-smoothed array and don't need much of a neighborhood
input_filter_start, input_filter_stop = roi.enlargeRoiForHalo(
output_start, output_stop, output_shape, 0.7, self.WINDOW_SIZE, enlarge_axes=axes2enlarge
)
# smooth roi in input frame
input_smooth_start, input_smooth_stop = roi.enlargeRoiForHalo(
input_filter_start,
input_filter_stop,
output_shape,
self.max_sigma,
self.WINDOW_SIZE,
enlarge_axes=axes2enlarge,
)
# target roi in filter frame
filter_target_start = roi.TinyVector(output_start - input_filter_start)
filter_target_stop = roi.TinyVector(output_stop - input_filter_start)
# filter roi in smooth frame
smooth_filter_start = roi.TinyVector(input_filter_start - input_smooth_start)
smooth_filter_stop = roi.TinyVector(input_filter_stop - input_smooth_start)
filter_target_slice = roi.roiToSlice(filter_target_start, filter_target_stop)
input_smooth_slice = roi.roiToSlice(input_smooth_start, input_smooth_stop)
# pre-smooth for all requested time slices and all channels
full_input_smooth_slice = (full_output_slice[0], slice(None), *input_smooth_slice)
req = self.Input[full_input_smooth_slice]
source = req.wait()
req.clean()
req.destination = None
if source.dtype != numpy.float32:
sourceF = source.astype(numpy.float32)
try:
source.resize((1,), refcheck=False)
except Exception:
pass
del source
source = sourceF
sourceV = source.view(vigra.VigraArray)
sourceV.axistags = copy.copy(self.Input.meta.axistags)
dimCol = len(self.scales)
dimRow = self.matrix.shape[0]
presmoothed_source = [None] * dimCol
source_smooth_shape = tuple(smooth_filter_stop - smooth_filter_start)
full_source_smooth_shape = (
full_output_stop[0] - full_output_start[0],
self.Input.meta.shape[1],
) + source_smooth_shape
try:
for j in range(dimCol):
for i in range(dimRow):
if self.matrix[i, j]:
# There is at least one filter op with this scale
break
else:
# There is no filter op at this scale
continue
if self.scales[j] > 1.0:
tempSigma = math.sqrt(self.scales[j] ** 2 - 1.0)
else:
tempSigma = self.scales[j]
presmoothed_source[j] = numpy.ndarray(full_source_smooth_shape, numpy.float32)
droi = (
(0, *tuple(smooth_filter_start._asint())),
(sourceV.shape[1], *tuple(smooth_filter_stop._asint())),
)
for i, vsa in enumerate(sourceV.timeIter()):
presmoothed_source[j][i, ...] = self._computeGaussianSmoothing(
vsa, tempSigma, droi, in2d=self.ComputeIn2d.value[j]
)
except RuntimeError as e:
if "kernel longer than line" in str(e):
raise RuntimeError(
"Feature computation error:\nYour image is too small to apply a filter with "
f"sigma={self.scales[j]:.1f}. Please select features with smaller sigmas."
)
else:
raise e
del sourceV
try:
source.resize((1,), refcheck=False)
except ValueError:
# Sometimes this fails, but that's okay.
logger.debug("Failed to free array memory.")
del source
cnt = 0
written = 0
closures = []
# connect individual operators
for i in range(dimRow):
for j in range(dimCol):
if self.matrix[i, j]:
oslot = self.featureOps[i][j].Output
req = None
slices = oslot.meta.shape[1]
if (
cnt + slices >= slot_roi.start[1]
and slot_roi.start[1] - cnt < slices
and slot_roi.start[1] + written < slot_roi.stop[1]
):
begin = 0
if cnt < slot_roi.start[1]:
begin = slot_roi.start[1] - cnt
end = slices
if cnt + end > slot_roi.stop[1]:
end = slot_roi.stop[1] - cnt
# feature slice in output frame
feature_slice = (slice(None), slice(written, written + end - begin)) + (slice(None),) * 3
subtarget = target[feature_slice]
# readjust the roi for the new source array
full_filter_target_slice = [full_output_slice[0], slice(begin, end), *filter_target_slice]
filter_target_roi = SubRegion(oslot, pslice=full_filter_target_slice)
closure = partial(
oslot.operator.execute,
oslot,
(),
filter_target_roi,
subtarget,
sourceArray=presmoothed_source[j],
)
closures.append(closure)
written += end - begin
cnt += slices
pool = RequestPool()
for c in closures:
pool.request(c)
pool.wait()
pool.clean()
for i in range(len(presmoothed_source)):
if presmoothed_source[i] is not None:
try:
presmoothed_source[i].resize((1,))
except Exception:
presmoothed_source[i] = None
def execute(self, slot, subindex, roi, result):
assert slot == self.Predictions or slot == self.Probabilities or slot == self.ProbabilityChannels
times = roi._l
if len(times) == 0:
# we assume that 0-length requests are requesting everything
times = range(self.Predictions.meta.shape[0])
forests=self.inputs["Classifier"][:].wait()
if forests is None or forests[0] is None:
# this happens if there was no data to train with
return dict((t, numpy.array([])) for t in times)
feats = {}
prob_predictions = {}
for t in times:
if t in self.prob_cache:
continue
tmpfeats = self.Features([t]).wait()
feats[t] = make_feature_array(tmpfeats)
prob_predictions[t] = [0] * len(forests)
def predict_forest(_t, forest_index):
# Note: We can't use RandomForest.predictLabels() here because we're training in parallel,
# and we have to average the PROBABILITIES from all forests.
# Averaging the label predictions from each forest is NOT equivalent.
# For details please see wikipedia:
# http://en.wikipedia.org/wiki/Electoral_College_%28United_States%29#Irrelevancy_of_national_popular_vote
# (^-^)
prob_predictions[_t][forest_index] = forests[forest_index].predictProbabilities(feats[_t])
# predict the data with all the forests in parallel
pool = RequestPool()
for t in times:
if t in self.prob_cache:
continue
for i, f in enumerate(forests):
req = Request( partial(predict_forest, t, i) )
pool.add(req)
pool.wait()
pool.clean()
for t in times:
if t not in self.prob_cache:
# prob_predictions is a dict-of-lists-of-arrays, indexed as follows:
# prob_predictions[t][forest_index][object_index, class_index]
# Stack the forests together and average them.
stacked_predictions = numpy.array( prob_predictions[t] )
averaged_predictions = numpy.average( stacked_predictions, axis=0 )
assert averaged_predictions.shape[0] == len(feats[t])
self.prob_cache[t] = averaged_predictions
self.prob_cache[t][0] = 0 # Background probability is always zero
if slot == self.Probabilities:
return { t : self.prob_cache[t] for t in times }
elif slot == self.Predictions:
# FIXME: Support SegmentationThreshold again...
labels = { t : 1 + numpy.argmax(self.prob_cache[t], axis=1) for t in times }
for t in times:
labels[t][0] = 0 # Background gets the zero label
return labels
elif slot == self.ProbabilityChannels:
prob_single_channel = {t: self.prob_cache[t][:, subindex[0]] for t in times}
return prob_single_channel
else:
assert False, "Unknown input slot"
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 extractHistograms(volume, labels, patchSize=64, haloSize=0,
nBins=30, intRange=(0, 255), appendPositions=False):
'''
extracts histograms from 3d-volume
- labels are
0 ignore
1 positive
2 negative
- histogram extraction is attempted to be done in parallel
- patches that intersect with the volume border are discarded
- volume and labels must be 3d, and in order 'zyx' (if not VigraArrays)
- returns: np.ndarray, shape: (nSamples,nBins+1), last column is the label
'''
# progress reporter class, histogram extraction can take quite a long time
class ProgressReporter(object):
lock = None
def __init__(self, nThreads):
self.lock = ThreadLock()
self.nThreads = nThreads
self.status = np.zeros((nThreads,))
def report(self, index):
self.lock.acquire()
self.status[index] = 1
logger.debug("Finished threads: %d/%d." %
(self.status.sum(), len(self.status)))
self.lock.release()
# sanity checks
assert len(volume.shape) == 3, "Volume must be 3d data"
assert volume.shape == labels.shape,\
"Volume and labels must have the same shape"
try:
volumeZYX = volume.withAxes(*'zyx')
labelsZYX = labels.withAxes(*'zyx')
except AttributeError:
# can't blame me
volumeZYX = volume
labelsZYX = labels
pass
# compute actual patch size
patchSize = patchSize + 2*haloSize
# fill list of patch centers (VigraArray does not support bitwise_or)
ind_z, ind_y, ind_x = np.where(
(labelsZYX == 1).view(np.ndarray) | (labelsZYX == 2).view(np.ndarray))
index = np.arange(len(ind_z))
# prepare chunking of histogram centers
chunkSize = 10000 # FIXME magic number??
nChunks = len(index)//chunkSize + (1 if len(index) % chunkSize > 0 else 0)
sliceList = [slice(k*chunkSize, min((k+1)*chunkSize, len(index)))
for k in range(nChunks)]
histoList = [None]*nChunks
# prepare subroutine for parallel extraction
reporter = ProgressReporter(nChunks)
#BEGIN subroutine
def _extractHistogramsSub(itemList):
xs = ind_x[itemList]
ys = ind_y[itemList]
zs = ind_z[itemList]
ymin = ys - patchSize//2
ymax = ymin + patchSize
xmin = xs - patchSize//2
xmax = xmin + patchSize
validPatchIndices = np.where(
np.all(
(ymin >= 0,
xmin >= 0,
xmax <= volumeZYX.shape[2],
ymax <= volumeZYX.shape[1]),
axis=0))[0]
if appendPositions:
out = np.zeros((len(validPatchIndices), nBins+4))
else:
out = np.zeros((len(validPatchIndices), nBins+1))
for k, patchInd in enumerate(validPatchIndices):
x = xs[patchInd]
y = ys[patchInd]
z = zs[patchInd]
vol = volumeZYX[z, ymin[patchInd]:ymax[patchInd],
xmin[patchInd]:xmax[patchInd]]
(out[k, :nBins], _) = np.histogram(
vol, bins=nBins, range=intRange, density=True)
out[k, nBins] = 1 if labelsZYX[z, y, x] == 1 else 0
if appendPositions:
out[k, nBins+1:] = [z, y, x]
return out
def partFun(i):
itemList = index[sliceList[i]]
histos = _extractHistogramsSub(itemList)
histoList[i] = histos
reporter.report(i)
#END subroutine
# pool the extraction requests
pool = RequestPool()
for i in range(nChunks):
req = Request(partial(partFun, i))
pool.add(req)
pool.wait()
pool.clean()
return np.vstack(histoList)