def testDdsCopyCreate(self):
"""
Test the :func:`mango.copy` method.
"""
dds0 = mango.ones(shape=self.shape, dtype="uint8")
dds1 = mango.copy(dds0, dtype="uint16")
self.assertEqual(dds0[0], dds1[0])
self.assertEqual(dds0[0, 0, 0], dds1[0, 0, 0])
self.assertEqual(dds0[(0, 0, 0)], dds1[(0, 0, 0)])
def testDdsSetGlobalOrigin(self):
"""
Test the :attr:`mango.Dds.origin` attribute for read/write access.
"""
origin = (5, 3, 1)
dds0 = mango.ones(shape=self.shape, dtype="uint8", origin=origin)
self.assertTrue(sp.all(origin == dds0.origin))
dds1 = mango.copy(dds0, dtype="uint16")
self.assertTrue(sp.all(origin == dds1.origin))
newOrigin = sp.array((8, 16, 2), dtype="int32")
dds0.origin = newOrigin
self.assertTrue(sp.all(newOrigin == dds0.origin))
self.assertTrue(
sp.all(dds1.subd.origin +
(newOrigin - origin) == dds0.subd.origin))
def testDdsResizeHalo(self):
for ddsH in [(0, 0, 0), (4, 4, 4), [2, 1, 0],
sp.array([3, 6, 7]), [6, 9, 8]]:
for h in [1, sp.array([1, 1, 1]), sp.array([2, 1, 0]), [3, 6, 7]]:
for s in [False, True]:
dds = \
mango.data.gaussian_noise(
shape=2*sp.array(self.shape),
mean = 32000.0,
stdd = 100.0,
mtype="tomo",
origin=(5,-11,7),
halo=ddsH
)
ddsResz = mango.copy(dds)
dds.updateHaloRegions()
dds.mirrorOuterLayersToBorder(True)
ddsResz.updateOverlapRegions()
ddsResz.mirrorOuterLayersToBorder(True)
ddsResz.resizeHalo(h, s)
if (not hasattr(h, "__len__")):
h = sp.array([
h,
] * 3)
h = sp.array(h)
self.assertListEqual(list(ddsResz.halo), list(h))
self.assertListEqual(list(dds.shape), list(ddsResz.shape))
self.assertListEqual(list(dds.origin),
list(ddsResz.origin))
self.assertListEqual(list(dds.subd.shape),
list(ddsResz.subd.shape))
self.assertListEqual(list(dds.subd.origin),
list(ddsResz.subd.origin))
self.assertListEqual(
list(dds.subd_h.shape - 2 * (dds.subd_h.halo - h)),
list(ddsResz.subd_h.shape))
self.assertListEqual(
list(dds.subd_h.origin + dds.subd_h.halo - h),
list(ddsResz.subd_h.origin))
self.assertTrue(
sp.all(dds.subd.asarray() == ddsResz.subd.asarray()))
slc = []
slcResz = []
for i in range(0, 3):
if (dds.halo[i] > h[i]):
slc.append(
slice(dds.halo[i] - h[i],
-(dds.halo[i] - h[i])))
slcResz.append(slice(None))
elif (dds.halo[i] < h[i]):
slc.append(slice(None))
slcResz.append(
slice(h[i] - dds.halo[i],
-(h[i] - dds.halo[i])))
else:
slc.append(slice(None))
slcResz.append(slice(None))
self.assertListEqual(
list(dds.subd_h.asarray()[slc].shape),
list(ddsResz.subd_h.asarray()[slcResz].shape))
self.assertTrue(
sp.all(dds.subd_h.asarray()[slc] ==
ddsResz.subd_h.asarray()[slcResz]))
def testDdsSetGlobalFaceValue(self):
"""
Test the :meth:`mango.Dds.setFacesToValue` method.
"""
outDir = self.createTmpDir("testDdsSetGlobalFaceValue")
ddsorig = mango.zeros(shape=self.shape,
mtype="tomo",
origin=(-8, 5, -32))
for depth in (1, int(np.min(ddsorig.shape) // 2)):
for axis in (0, 1, 2):
dds = mango.copy(ddsorig)
vallo = 16 + axis
rootLogger.info("LO: axis=%s, val=%s, depth=%s" %
(axis, vallo, depth))
dds.setFacesToValue(vallo, axislo=axis, depth=depth)
#mango.io.writeDds(os.path.join(outDir, "tomoLoFaceAxis%s.nc" % axis), dds)
shpFace = dds.shape
shpFace[axis] = depth
mpidimsFace = [0, 0, 0]
mpidimsFace[axis] = 1
rootLogger.info("LO: shpFace=%s, mpidimsFace=%s" %
(shpFace, mpidimsFace))
ddsFace = mango.copy(dds, shape=shpFace, mpidims=mpidimsFace)
self.assertTrue(sp.all(ddsFace.subd.asarray() == vallo))
shpNotFace = dds.shape
shpNotFace[axis] = dds.shape[axis] - depth
orgNotFace = dds.origin
orgNotFace[axis] = dds.origin[axis] + depth
mpidimsNotFace = [0, 0, 0]
mpidimsNotFace[axis] = 1
rootLogger.info(
"LO: shpNotFace=%s, mpidimsNotFace=%s, orgNotFace=%s" %
(shpNotFace, mpidimsNotFace, orgNotFace))
ddsNotFace = mango.copy(dds,
origin=orgNotFace,
shape=shpNotFace,
mpidims=mpidimsNotFace)
self.assertTrue(sp.all(ddsNotFace.subd.asarray() == 0))
dds = mango.copy(ddsorig)
valhi = 2 * vallo
rootLogger.info("HI: axis=%s, val=%s, depth=%s" %
(axis, valhi, depth))
dds.setFacesToValue(valhi, axishi=axis, depth=depth)
#mango.io.writeDds(os.path.join(outDir, "tomoLoFaceAxis%s.nc" % axis), dds)
shpFace = dds.shape
shpFace[axis] = depth
mpidimsFace = [0, 0, 0]
mpidimsFace[axis] = 1
orgFace = dds.origin
orgFace[axis] = dds.origin[axis] + dds.shape[axis] - depth
rootLogger.info("HI: shpFace=%s, mpidimsFace=%s, orgFace=%s" %
(shpFace, mpidimsFace, orgFace))
ddsFace = mango.copy(dds,
origin=orgFace,
shape=shpFace,
mpidims=mpidimsFace)
self.assertTrue(sp.all(ddsFace.subd.asarray() == valhi))
shpNotFace = dds.shape
shpNotFace[axis] = dds.shape[axis] - depth
mpidimsNotFace = [0, 0, 0]
mpidimsNotFace[axis] = 1
rootLogger.info(
"HI: shpNotFace=%s, mpidimsNotFace=%s, orgNotFace=%s" %
(shpNotFace, mpidimsNotFace, orgNotFace))
ddsNotFace = mango.copy(dds,
shape=shpNotFace,
mpidims=mpidimsNotFace)
self.assertTrue(sp.all(ddsNotFace.subd.asarray() == 0))
def calcSphericalCavityLabels(self, dds):
mpidims = dds.mpi.shape
if (dds.mpi.comm != None):
mpidims = (dds.mpi.comm.Get_size(), 1, 1)
dds = mango.copy(dds, mpidims=mpidims)
rootLogger.info("Calculating neighbourhood mean image...")
fltDds = mango.copy(dds,
mtype="tomo_float",
halo=self.se.getHaloSize())
nMeanDds = mango.image.mean_filter(fltDds, self.se)
nMeanDds = mango.copy(nMeanDds, halo=(0, 0, 0))
rootLogger.info("Calculating neighbourhood stdd image...")
nStddDds = mango.image.stdd_filter(fltDds, self.se)
nStddDds = mango.copy(nStddDds, halo=(0, 0, 0))
del fltDds
self.maskLowStddVoxels(dds, nMeanDds, nStddDds)
rootLogger.info("Calculating mean vs stdd histogram...")
h2d, edges = mango.image.histogramdd([nMeanDds, nStddDds],
bins=(1024, 8192))
h2d = h2d[:, :-2]
rootLogger.info(
"h2d.shape = %s, edges[0].shape=%s, edges[1].shape=%s" %
(h2d.shape, edges[0].shape, edges[1].shape))
rootLogger.info("Done calculating mean vs stdd histogram...")
maxIdx = np.unravel_index(np.argmax(h2d), h2d.shape)
rootLogger.info("np.argmax(h2d) = %s" % (maxIdx, ))
backgroundMean = 0.5 * (edges[0][maxIdx[0]] + edges[0][maxIdx[0] + 1])
backgroundStdd = 0.5 * (edges[1][maxIdx[1]] + edges[1][maxIdx[1] + 1])
mskDds = mango.copy(dds)
rootLogger.info("Background (mean,stdd) = (%s, %s)." %
(backgroundMean, backgroundStdd))
mskDds.subd.asarray()[...] = \
sp.where(
sp.logical_or(
nStddDds.subd.asarray() < (8 * backgroundStdd),
nMeanDds.subd.asarray() < (backgroundMean + 3 * backgroundStdd),
),
mskDds.mtype.maskValue(),
mskDds.subd.asarray()
)
del nMeanDds, nStddDds
self.writeIntermediateDds("_AaaPrePercentileTailMask", mskDds)
self.eliminatePercentileTails(mskDds, 1.0, 92.5)
rootLogger.info("Calculating neighbourhood stdd image...")
nStddDds = mango.image.stdd_filter(mskDds, self.se)
self.eliminatePercentileTails(nStddDds, 33.0, 100.0)
rootLogger.info("Copying stdd percentile tail mask to mskDds...")
mango.copy_masked(nStddDds, mskDds)
rootLogger.info("Done copying stdd percentile tail mask to mskDds.")
self.writeIntermediateDds("_AaaPstPercentileTailMask", mskDds)
rootLogger.info("Eliminating small clusters from mskDds...")
self.eliminateSmallClusters(mskDds, 0.1)
rootLogger.info("Done eliminating small clusters from mskDds.")
self.writeIntermediateDds("_AaaPstSmallClusterMask", mskDds)
del nStddDds
segEdtDds = mango.zeros_like(mskDds, mtype="segmented")
segEdtDds.asarray()[...] = sp.where(
mskDds.asarray() == mskDds.mtype.maskValue(), 0, 1)
self.writeIntermediateDds("_AaaPreCvxHullMask", segEdtDds)
rootLogger.info("Calculating convex hull...")
cvxHullMsk = mango.image.convex_hull_3d(
segEdtDds,
inputmsk=0,
outhull=segEdtDds.mtype.maskValue(),
inhull=1)
segEdtDds.asarray()[...] = sp.where(
cvxHullMsk.asarray() == cvxHullMsk.mtype.maskValue(), 1,
segEdtDds.asarray())
rootLogger.info("Done calculating convex hull.")
self.writeIntermediateDds("_AaaPstCvxHullMask", segEdtDds)
segEdtDds.setFacesToValue(1)
rootLogger.info("Calculating EDT image...")
edtDds = mango.image.distance_transform_edt(segEdtDds, val=0)
self.writeIntermediateDds("_AaaPstCvxHullMaskEdt", edtDds)
rootLogger.info("Calculating MCR image...")
mcrDds = mango.image.max_covering_radius(edtDds,
maxdist=0.5 *
(np.min(edtDds.shape)),
filecache=True)
mango.copy_masked(cvxHullMsk, mcrDds)
rootLogger.info("Calculating (min,max) MCR values...")
mcrMin, mcrMax = mango.minmax(mcrDds)
rootLogger.info("Masking small MCR values...")
mcrDds.asarray()[...] = sp.where(mcrDds.asarray() >= 0.05 * mcrMax,
mcrDds.asarray(),
mcrDds.mtype.maskValue())
self.writeIntermediateDds("_AaaPstCvxHullMaskMcr", mcrDds)
del cvxHullMsk, edtDds
#
# Normalise the intensities so that a relative-gradient is computed for the largest
# MCR radii.
rootLogger.info("Normalising MCR image...")
mnmx = mango.minmax(mcrDds)
tmpDds = mango.copy(mcrDds)
tmpDds.asarray()[...] -= mnmx[1]
tmpDds.asarray()[...] *= tmpDds.asarray()
tmpDds.asarray()[...] = 1 + mnmx[1] * np.exp(-(tmpDds.asarray()) /
(2 * 0.133 * 0.133 *
(mnmx[1] * mnmx[1])))
mcrDds.asarray()[...] = sp.where(mcrDds.asarray() > 0,
mcrDds.asarray() / tmpDds.asarray(),
mcrDds.asarray())
rootLogger.info("Calculating MCR image gradient...")
grdMcrDds = mango.image.discrete_gaussian_gradient_magnitude(
mcrDds, 0.65, errtol=0.01)
grdMcrDds.asarray()[...] = sp.where(grdMcrDds.asarray() <= 3.0e-2,
mcrDds.asarray(),
mcrDds.mtype.maskValue())
rootLogger.info("Calculating unique MCR low-gradient values...")
u = mango.unique(grdMcrDds)
rootLogger.info(
"Converting low gradient MCR image to binary segmentation...")
segDds = mango.map_element_values(grdMcrDds,
lambda x: x in u,
mtype="segmented")
rootLogger.info("Labeling low gradient MCR regions...")
mango.copy_masked(mcrDds, segDds)
lblDds = mango.image.label(segDds, val=1, connectivity=26, dosort=True)
self.writeIntermediateDds("_AaaPstCvxHullMaskMcrGrdLbl", lblDds)
del segDds, grdMcrDds
rootLogger.info("Calculating Principal Moment of Inertia...")
self.pmoi, self.pmoi_axes, self.com = mango.image.moment_of_inertia(
mskDds)
rootLogger.info("Done calculating Principal Moment of Inertia.")
return lblDds, mcrDds, segEdtDds
def calcSphericalCavityLabels(self, dds):
mpidims = dds.mpi.shape
if (dds.mpi.comm != None):
mpidims = (dds.mpi.comm.Get_size(), 1, 1)
dds = mango.copy(dds, mpidims=mpidims)
rootLogger.info("Calculating neighbourhood mean image...")
fltDds = mango.copy(dds, mtype="tomo_float", halo=self.se.getHaloSize())
nMeanDds = mango.image.mean_filter(fltDds, self.se)
nMeanDds = mango.copy(nMeanDds, halo=(0,0,0))
rootLogger.info("Calculating neighbourhood stdd image...")
nStddDds = mango.image.stdd_filter(fltDds, self.se)
nStddDds = mango.copy(nStddDds, halo=(0,0,0))
del fltDds
self.maskLowStddVoxels(dds, nMeanDds, nStddDds)
rootLogger.info("Calculating mean vs stdd histogram...")
h2d, edges = mango.image.histogramdd([nMeanDds, nStddDds], bins=(1024, 8192))
h2d = h2d[:, :-2]
rootLogger.info("h2d.shape = %s, edges[0].shape=%s, edges[1].shape=%s" % (h2d.shape, edges[0].shape, edges[1].shape))
rootLogger.info("Done calculating mean vs stdd histogram...")
maxIdx = np.unravel_index(np.argmax(h2d), h2d.shape)
rootLogger.info("np.argmax(h2d) = %s" % (maxIdx,))
backgroundMean = 0.5 * (edges[0][maxIdx[0]] + edges[0][maxIdx[0] + 1])
backgroundStdd = 0.5 * (edges[1][maxIdx[1]] + edges[1][maxIdx[1] + 1])
mskDds = mango.copy(dds)
rootLogger.info("Background (mean,stdd) = (%s, %s)." % (backgroundMean, backgroundStdd))
mskDds.subd.asarray()[...] = \
sp.where(
sp.logical_or(
nStddDds.subd.asarray() < (8 * backgroundStdd),
nMeanDds.subd.asarray() < (backgroundMean + 3 * backgroundStdd),
),
mskDds.mtype.maskValue(),
mskDds.subd.asarray()
)
del nMeanDds, nStddDds
self.writeIntermediateDds("_AaaPrePercentileTailMask", mskDds)
self.eliminatePercentileTails(mskDds, 1.0, 92.5)
rootLogger.info("Calculating neighbourhood stdd image...")
nStddDds = mango.image.stdd_filter(mskDds, self.se)
self.eliminatePercentileTails(nStddDds, 33.0, 100.0)
rootLogger.info("Copying stdd percentile tail mask to mskDds...")
mango.copy_masked(nStddDds, mskDds)
rootLogger.info("Done copying stdd percentile tail mask to mskDds.")
self.writeIntermediateDds("_AaaPstPercentileTailMask", mskDds)
rootLogger.info("Eliminating small clusters from mskDds...")
self.eliminateSmallClusters(mskDds, 0.1)
rootLogger.info("Done eliminating small clusters from mskDds.")
self.writeIntermediateDds("_AaaPstSmallClusterMask", mskDds)
del nStddDds
segEdtDds = mango.zeros_like(mskDds, mtype="segmented")
segEdtDds.asarray()[...] = sp.where(mskDds.asarray() == mskDds.mtype.maskValue(), 0, 1)
self.writeIntermediateDds("_AaaPreCvxHullMask", segEdtDds)
rootLogger.info("Calculating convex hull...")
cvxHullMsk = mango.image.convex_hull_3d(segEdtDds, inputmsk=0, outhull=segEdtDds.mtype.maskValue(), inhull=1)
segEdtDds.asarray()[...] = sp.where(cvxHullMsk.asarray() == cvxHullMsk.mtype.maskValue(), 1, segEdtDds.asarray())
rootLogger.info("Done calculating convex hull.")
self.writeIntermediateDds("_AaaPstCvxHullMask", segEdtDds)
segEdtDds.setFacesToValue(1)
rootLogger.info("Calculating EDT image...")
edtDds = mango.image.distance_transform_edt(segEdtDds, val=0)
self.writeIntermediateDds("_AaaPstCvxHullMaskEdt", edtDds)
rootLogger.info("Calculating MCR image...")
mcrDds = mango.image.max_covering_radius(edtDds, maxdist=0.5*(np.min(edtDds.shape)), filecache=True)
mango.copy_masked(cvxHullMsk, mcrDds)
rootLogger.info("Calculating (min,max) MCR values...")
mcrMin, mcrMax = mango.minmax(mcrDds)
rootLogger.info("Masking small MCR values...")
mcrDds.asarray()[...] = sp.where(mcrDds.asarray() >= 0.05*mcrMax, mcrDds.asarray(), mcrDds.mtype.maskValue())
self.writeIntermediateDds("_AaaPstCvxHullMaskMcr", mcrDds)
del cvxHullMsk, edtDds
#
# Normalise the intensities so that a relative-gradient is computed for the largest
# MCR radii.
rootLogger.info("Normalising MCR image...")
mnmx = mango.minmax(mcrDds)
tmpDds = mango.copy(mcrDds)
tmpDds.asarray()[...] -= mnmx[1]
tmpDds.asarray()[...] *= tmpDds.asarray()
tmpDds.asarray()[...] = 1+mnmx[1]*np.exp(-(tmpDds.asarray())/(2*0.133*0.133*(mnmx[1]*mnmx[1])))
mcrDds.asarray()[...] = sp.where(mcrDds.asarray() > 0, mcrDds.asarray()/tmpDds.asarray(), mcrDds.asarray())
rootLogger.info("Calculating MCR image gradient...")
grdMcrDds = mango.image.discrete_gaussian_gradient_magnitude(mcrDds, 0.65, errtol=0.01)
grdMcrDds.asarray()[...] = sp.where(grdMcrDds.asarray() <= 3.0e-2, mcrDds.asarray(), mcrDds.mtype.maskValue())
rootLogger.info("Calculating unique MCR low-gradient values...")
u = mango.unique(grdMcrDds)
rootLogger.info("Converting low gradient MCR image to binary segmentation...")
segDds = mango.map_element_values(grdMcrDds, lambda x: x in u, mtype="segmented")
rootLogger.info("Labeling low gradient MCR regions...")
mango.copy_masked(mcrDds, segDds)
lblDds = mango.image.label(segDds, val=1, connectivity=26, dosort=True)
self.writeIntermediateDds("_AaaPstCvxHullMaskMcrGrdLbl", lblDds)
del segDds, grdMcrDds
rootLogger.info("Calculating Principal Moment of Inertia...")
self.pmoi, self.pmoi_axes, self.com = mango.image.moment_of_inertia(mskDds)
rootLogger.info("Done calculating Principal Moment of Inertia.")
return lblDds, mcrDds, segEdtDds
def cylinder_fit_multi_res(
input,
numcyl,
retcylimg=False,
resolutions=[128, 256, 1024],
metric = None,
radius_range_percent = (50.0, 100.0),
centre_range_percent = (-8.0, 8.0),
metricThreshRelTol = 0.25
):
"""
Multi-resolution version of :func:`cylinder_fit`.
:type input: :obj:`mango.Dds`
:param input: Search for cylinder boundaries in the gradient-vector
image of this :samp:`{input}` image.
:type numcyl: :obj:`mango.Dds`
:param numcyl: Number of cylinders to find.
:type retcylimg: :obj:`bool`
:param retcylimg: If :samp:`True`, additionally returns a segmented image
with the boundary of the cylinder fits labeled from :samp:`0` to :samp:`{numcyl}-1`.
:type resolutions: sequence of :obj:`int`
:param resolutions: The sequence of image shapes for each cylinder fit resolution.
Exhaustive search and refinement is performed at the lowest resolution,
just refinement at the higher resolutions.
:type metric: :obj:`CylinderFitGradientMetric`
:param metric: Metric object used to identify cylinder parameters
which indicate the presence of a cylindrical boundary in the :samp:`input` image.
:type radius_range_percent: 2 or 3 sequence of :obj:`float`
:param radius_range_percent: Indicates the cylinder-radius exhaustive search grid
as :samp:`(min_percent, max_percent, percent_step)`. This is a percentage
of the minimum :samp:`{input}.shape` component (i.e. :samp:`numpy.min({input}.shape)`).
If the percent step size is omitted, then a percent equivalent of 2-voxels is chosen
as the step-size.
:type centre_range_percent: 2 or 3 sequence of :obj:`float`
:param centre_range_percent: Indicates the cylinder-centre exhaustive search grid
as :samp:`(min_percent, max_percent, percent_step)`. This is a percentage
of the minimum :samp:`{input}.shape` component (i.e. :samp:`numpy.min({input}.shape)`).
If the percent step size is omitted, then a percent equivalent of 2-voxels is chosen
as the step-size.
:type metricThreshRelTol: :obj:`float`
:param metricThreshRelTol: Relative tolerance for truncating the
exhaustive search metric values. Ignore exhaustive search parameters
which give metric values that have relative difference greater
than :samp:`{metricThreshRelTol}` of the best (lowest) metric value in
the entire search.
:rtype: :obj:`list` of :samp:`[metricVal, cylinderParameters]` pairs
:return: List of pairs :samp:`pairList = [[metricVal0, cylPrms0], [metricVal1, cylPrms1], ...]`
where :samp:`cylPrms0=[[centrez0, centrey0, centrex0], radius0, axisLength0, [axisz0, axisy0, axisx0]]`, :samp:`cylPrms1=[[centrez1, centrey1, centrex1], radius1, axisLength1, [axisz1, axisy1, axisx1]]`, etc.
If :samp:`{retcylimg}` is :samp:`True`, additionally returns segmented :obj:`mango.Dds`
image labeled with cylinder boundaries (i.e. returns :samp:`pairList, segDds`).
"""
resolutions = list(resolutions)
resolutions.sort()
dsFactors = np.min(input.shape)/sp.array(resolutions, dtype="float64")
if (dsFactors[-1] < 1):
dsFactors[-1] = 1.0
dsFactors = sp.array([dsFactor for dsFactor in dsFactors if dsFactor >= 1], dtype=dsFactors.dtype)
i = 0
rootLogger.info("Downsample factors = %s." % (dsFactors,))
rootLogger.info("Downsampling (resolution=%s) input.shape=%s to shape=%s..." % (resolutions[i], input.shape, input.shape/dsFactors[i]))
input.updateOverlapRegions()
input.mirrorOuterLayersToBorder(False)
dsInput = mango.image.gaussian_downsample(input, factor=[1.0/dsFactors[0],]*3)
dsInput.updateHaloRegions()
dsInput.mirrorOuterLayersToBorder(False)
# Do data-parallel parallelism for all resolutions, having a copy of
# the global downsampled gradient-vector image on all processes tends
# to be too memory hungary.
distributedMetricEvaluation = True
rootLogger.info("Done downsampling, new shape=%s." % (dsInput.shape, ))
cylPrmList = \
cylinder_fit(
input = dsInput,
numcyl=numcyl,
retcylimg = False,
distributedMetricEvaluation = distributedMetricEvaluation,
metric = metric,
radius_range_percent = radius_range_percent,
centre_range_percent = centre_range_percent,
metricThreshRelTol = metricThreshRelTol
)
rootLogger.info("Lowest resolution fit cylinders=:\n%s" % ("\n".join(map(str, cylPrmList))))
del dsInput
dsInput = None
fitter = None
distributedMetricEvaluation = True
i = 1
if (metric == None):
metric = cylinder_fit_gradient_metric()
for dsFactor in dsFactors[1:]:
metric.setNumSlices(metric.getNumSlices() + 10)
metric.setNumPointsPerSlice(int(metric.getNumPointsPerSlice()*1.25))
if (dsFactor > 1):
rootLogger.info("Downsampling (resolution=%s) input.shape=%s to shape=%s..." % (resolutions[i], input.shape, input.shape/dsFactor))
dsInput = mango.image.gaussian_downsample(input, factor=[1.0/dsFactor,]*3)
rootLogger.info("Done downsampling, new shape=%s." % (dsInput.shape, ))
else:
rootLogger.info("Full resolution refinement input.shape=%s..." % (input.shape,))
dsInput = input
if (input.mpi.comm != None):
input.mpi.comm.barrier()
dsInput.updateHaloRegions()
dsInput.mirrorOuterLayersToBorder(False)
rootLogger.info("Cylinder-fit refinement for resolution image shape=%s." % (dsInput.shape,))
fitter = \
_CylinderFitter(
input = dsInput,
numcyl = numcyl,
distributedMetricEvaluation = distributedMetricEvaluation,
metric = metric,
radius_range_percent = radius_range_percent,
centre_range_percent = centre_range_percent,
axis_range_degrees = None,
metricThreshRelTol = metricThreshRelTol
)
if (i >= (len(dsFactors)-1)):
fitter.minimize_options['xtol'] = 1.0e-4
fitter.minimize_options['ftol'] = (fitter.minimize_options['xtol'])**2
cylPrmList = fitter.refineFit(cylPrmList)
rootLogger.info("Done cylinder-fit refinement for resolution image shape=%s." % (dsInput.shape,))
rootLogger.info("Fit cylinders=:\n%s" % ("\n".join(map(str, cylPrmList))))
i+=1
del dsInput, fitter
dsInput = None
fitter = None
retVal = cylPrmList
if (retcylimg):
cylPrmList = [cylfit[1] for cylfit in cylPrmList]
cylImg = mango.copy(input, mtype="segmented")
cylImg.setAllToValue(cylImg.mtype.maskValue())
cylImg = fill_cylinder_boundaries(cylPrmList, cylImg)
retVal = [retVal, cylImg]
return retVal