def test_GroupAdjointAction(self, disp=False):
hM = common.RandField(self.sz,
nSig=5.0,
gSig=4.0,
mType=ca.MEM_HOST,
sp=self.imSp)
hV = common.RandField(self.sz,
nSig=5.0,
gSig=4.0,
mType=ca.MEM_HOST,
sp=self.imSp)
hPhi = common.RandField(self.sz,
nSig=5.0,
gSig=4.0,
mType=ca.MEM_HOST,
sp=self.imSp)
tmp = ca.Field3D(self.grid, ca.MEM_HOST)
# compute < m, Ad_\phi v >
ca.Ad(tmp, hPhi, hV)
rhs = ca.Dot(tmp, hM)
# compute < Ad^*_\phi m, v >
ca.CoAd(tmp, hPhi, hM)
lhs = ca.Dot(tmp, hV)
#print "a=%f b=%f" % (rhs, lhs)
self.assertLess(abs(rhs - lhs), 2e-6)
def predict_each_datapart(args, net, network_config, input_batch, datapart_idx, batch_size, patch_size, predict_transform_space):
moving_image = torch.load(args.moving_image_dataset[datapart_idx])
target_image = torch.load(args.target_image_dataset[datapart_idx])
optimization_momentum = torch.load(args.deformation_parameter[datapart_idx])
for slice_idx in range(0, moving_image.size()[0]):
print(slice_idx)
moving_slice = moving_image[slice_idx].numpy()
target_slice = target_image[slice_idx].numpy()
if predict_transform_space:
moving_slice = util.convert_to_registration_space(moving_slice)
target_slice = util.convert_to_registration_space(target_slice)
predicted_momentum = util.predict_momentum(moving_slice, target_slice, input_batch, batch_size, patch_size, net, predict_transform_space);
m0_reg = common.FieldFromNPArr(predicted_momentum['image_space'], ca.MEM_DEVICE);
moving_image_ca = common.ImFromNPArr(moving_slice, ca.MEM_DEVICE)
target_image_ca = common.ImFromNPArr(target_slice, ca.MEM_DEVICE)
registration_result = registration_methods.geodesic_shooting(moving_image_ca, target_image_ca, m0_reg, args.shoot_steps, ca.MEM_DEVICE, network_config)
target_inv = common.AsNPCopy(registration_result['I1_inv'])
print(target_inv.shape)
if predict_transform_space:
target_inv = util.convert_to_predict_space(target_inv)
print(target_inv.shape)
target_inv = torch.from_numpy(target_inv)
target_image[slice_idx] = target_inv
optimization_momentum[slice_idx] = optimization_momentum[slice_idx] - torch.from_numpy(predicted_momentum['prediction_space'])
torch.save(target_image, args.warped_back_target_output[datapart_idx])
torch.save(optimization_momentum, args.momentum_residual[datapart_idx])
def test_SplatWorldAdjoint(self, disp=False):
# Random input images
reg = common.RandImage(self.sz,
nSig=1.0,
gSig=0.0,
mType=ca.MEM_HOST,
sp=self.imSp)
# adjust hJ's grid so that voxels don't align perfectly
spfactor = 0.77382
small = common.RandImage(self.sz,
nSig=1.0,
gSig=0.0,
mType=ca.MEM_HOST,
sp=self.imSp * spfactor)
tmpsmall = ca.Image3D(small.grid(), ca.MEM_HOST)
ca.SetMem(tmpsmall, 0)
# compute < I(Phi(x)), J(x) >
ca.ResampleWorld(tmpsmall, reg)
smallIP = ca.Dot(tmpsmall, small)
# compute < I(y), |DPhi^{-1}(y)| J(Phi^{-1}(y)) >
tmpreg = ca.Image3D(reg.grid(), ca.MEM_HOST)
ca.SetMem(tmpreg, 0)
ca.SplatWorld(tmpreg, small)
regIP = ca.Dot(tmpreg, reg)
#print "a=%f b=%f" % (phiIdotJ, IdotphiJ)
self.assertLess(abs(smallIP - regIP), 2e-6)
def gather_file(args):
if args.momentum:
file = common.AsNPCopy(common.LoadITKField(args.files[0], ca.MEM_HOST))
else:
#file = common.AsNPCopy(common.LoadITKImage(args.files[0], ca.MEM_HOST))
file = nib.load(args.files[0]).get_data()
all_size = (len(args.files), ) + (15, 15, 15, 1, 3) #file.shape;
data = torch.zeros(all_size)
for i in range(0, len(args.files)):
if args.momentum:
cur_slice = torch.from_numpy(
common.AsNPCopy(common.LoadITKField(args.files[i],
ca.MEM_HOST)))
else:
cur_slice = nib.load(args.files[i]).get_data()
if cur_slice.size == 3375:
cur_slice = np.zeros([15, 15, 15, 1, 3])
cur_slice = torch.from_numpy(cur_slice)
data[i] = cur_slice
if args.momentum:
# transpose the dataset to fit the training format
data = data.numpy()
data = np.transpose(data, [0, 4, 1, 2, 3])
data = torch.from_numpy(data)
torch.save(data, args.output)
def test_SplatAdjoint(self, disp=False):
hI = common.RandImage(self.sz,
nSig=1.0,
gSig=0.0,
mType=ca.MEM_HOST,
sp=self.imSp)
hJ = common.RandImage(self.sz,
nSig=1.0,
gSig=0.0,
mType=ca.MEM_HOST,
sp=self.imSp)
hPhi = common.RandField(self.sz,
nSig=5.0,
gSig=4.0,
mType=ca.MEM_HOST,
sp=self.imSp)
tmp = ca.Image3D(self.grid, ca.MEM_HOST)
# compute < I(Phi(x)), J(x) >
ca.ApplyH(tmp, hI, hPhi)
phiIdotJ = ca.Dot(tmp, hJ)
# compute < I(y), |DPhi^{-1}(y)| J(Phi^{-1}(y)) >
ca.Splat(tmp, hPhi, hJ)
IdotphiJ = ca.Dot(tmp, hI)
#print "a=%f b=%f" % (phiIdotJ, IdotphiJ)
self.assertLess(abs(phiIdotJ - IdotphiJ), 2e-6)
def AtlasWriteOutput(cf, atlas, m_array, nodeSubjectsIds, isReporter):
# save initial momenta for all individuals
for itsub in range(len(nodeSubjectsIds)):
common.SaveITKField(
m_array[itsub], cf.io.outputPrefix +
str(nodeSubjectsIds[itsub]).replace('.', '_') + "_m0.mhd")
# save the atlas
if isReporter:
common.SaveITKImage(atlas, cf.io.outputPrefix + "atlas.mhd")
def GeoRegWriteOuput(subjectId, cf, p, t, Imsmts, cpinds, cpstates, msmtinds,
gradAtMsmts, EnergyHistory):
# save initial image and momenta for regression geodesic
common.SaveITKImage(p.I0, cf.io.outputPrefix + subjectId + "I0.mhd")
common.SaveITKField(p.m0, cf.io.outputPrefix + subjectId + "m0.mhd")
# save residual images for regression geodesic
# TODO:
# write Energy details
energyFilename = cf.io.outputPrefix + subjectId + "Energy.csv"
with open(energyFilename, 'w') as f:
csv_writer = csv.writer(f, delimiter='\t')
csv_writer.writerows(EnergyHistory)
def MyQuiver(vf, sliceIdx=None, dim='z',every=1,thresh=None,scaleArrows=None,arrowCol='r',lineWidth=None, width=None):
sliceArr = common.ExtractSliceArrVF(vf, sliceIdx, dim)
if dim=='z':
vy = np.squeeze(sliceArr[:,:,0])
vx = np.squeeze(sliceArr[:,:,1])
elif dim=='x':
vy = np.squeeze(sliceArr[:,:,1])
vx = np.squeeze(sliceArr[:,:,2])
elif dim=='y':
vy = np.squeeze(sliceArr[:,:,0])
vx = np.squeeze(sliceArr[:,:,2])
vxshow = np.zeros(np.shape(vx))
vyshow = np.zeros(np.shape(vy))
vxshow[::every,::every] = vx[::every,::every]
vyshow[::every,::every] = vy[::every,::every]
valindex = np.zeros(np.shape(vx),dtype=bool)
valindex[::every,::every] = True
if thresh is not None:
valindex[(vx**2+vy**2)<thresh] = False
#gridX, gridY = np.meshgrid(range(np.shape(vx)[0]),range(np.shape(vx)[1]-1,-1,-1))
gridX, gridY = np.meshgrid(range(np.shape(vx)[0]),range(np.shape(vx)[1]))
quiverhandle = plt.quiver(gridX[valindex],gridY[valindex],vx[valindex],vy[valindex],scale=scaleArrows,color=arrowCol,linewidth=lineWidth,width=width,zorder=4)
# change axes to match image axes
plt.gca().invert_yaxis()
# force redraw
plt.draw()
def intensity_normalization_histeq(args):
for i in range(0, len(args.input_images)):
image = common.LoadITKImage(args.output_images[i], ca.MEM_HOST)
grid = image.grid()
image_np = common.AsNPCopy(image)
nan_mask = np.isnan(image_np)
image_np[nan_mask] = 0
image_np /= np.amax(image_np)
# perform histogram equalization if needed
if args.histeq:
image_np[image_np != 0] = exposure.equalize_hist(
image_np[image_np != 0])
image_result = common.ImFromNPArr(image_np, ca.MEM_HOST)
image_result.setGrid(grid)
common.SaveITKImage(image_result, args.output_images[i])
def randHSetUp(self):
self.hRandH = \
common.RandField(self.sz, nSig=5.0, gSig=4.0,
mType = MEM_HOST, sp = self.imSp)
VtoH_I(self.hRandH)
self.dRandH = self.hRandH.copy()
self.dRandH.toType(MEM_DEVICE)
def GridPlot(vf, sliceIdx=None, dim='z', every=1,
isVF=True, color='g', plotBase=True, colorbase='#A0A0FF'):
sliceArr = np.squeeze(common.ExtractSliceArrVF(vf, sliceIdx, dim))
sz = sliceArr.shape
hID = np.mgrid[0:sz[0], 0:sz[1]]
d1 = np.squeeze(hID[1, ::every, ::every])
d2 = np.squeeze(hID[0, ::every, ::every])
sliceArr = sliceArr[::every, ::every, :]
if plotBase:
plt.plot(d1, d2, colorbase)
plt.hold(True)
plt.plot(d1.T, d2.T, colorbase)
if not isVF:
d1 = np.zeros(d1.shape)
d2 = np.zeros(d2.shape)
plt.plot(d1+np.squeeze(sliceArr[:,:,1]),
d2+np.squeeze(sliceArr[:,:,0]), color)
plt.hold(True)
plt.plot((d1+np.squeeze(sliceArr[:,:,1])).T,
(d2+np.squeeze(sliceArr[:,:,0])).T, color)
plt.hold(False)
# change axes to match image axes
if not plt.gca().yaxis_inverted():
plt.gca().invert_yaxis()
# force redraw
plt.draw()
def randVPair(self):
hV = common.RandField(self.sz,
nSig=5.0,
gSig=4.0,
mType=MEM_HOST,
sp=self.imSp)
dV = hV.copy()
dV.toType(MEM_DEVICE)
return hV, dV
def randMaskSetUp(self):
randArr = np.random.rand(self.sz[0], self.sz[1])
maskArr = np.zeros(randArr.shape)
maskArr[randArr > 0.5] = 1.0
self.hRandMask = common.ImFromNPArr(maskArr,
mType=MEM_HOST,
sp=self.imSp)
self.dRandMask = self.hRandMask.copy()
self.dRandMask.toType(MEM_DEVICE)
def test_ResampleInterp(self, disp=False):
# generate small integer-valued image
initMax = 5
randArrSmall = (np.random.rand(10, 10) * initMax).astype(int)
randImSmall = common.ImFromNPArr(randArrSmall)
imLarge = Image3D(50, 50, 1)
Resample(imLarge, randImSmall, BACKGROUND_STRATEGY_CLAMP, INTERP_NN)
nUnique = len(np.unique(imLarge.asnp()))
self.assertEqual(nUnique, initMax)
def Quiver(vf, sliceIdx=None, dim='z'):
sliceArr = common.ExtractSliceArrVF(vf, sliceIdx, dim)
plt.quiver(np.squeeze(sliceArr[:,:,0]).T,
np.squeeze(sliceArr[:,:,1]).T,
scale=None)
# change axes to match image axes
if not plt.gca().yaxis_inverted():
plt.gca().invert_yaxis()
# force redraw
plt.draw()
def main():
secNum = sys.argv[1]
mkyNum = sys.argv[2]
channel = sys.argv[3]
region = str(sys.argv[4])
conf_dir = '/home/sci/blakez/korenbergNAS/3D_database/Working/Microscopic/confocal/src_registration/'
side_dir = '/home/sci/blakez/korenbergNAS/3D_database/Working/Microscopic/side_light_microscope/src_registration/'
save_dir = '/home/sci/blakez/korenbergNAS/3D_database/Working/Microscopic/confocal/sidelight_registered/'
# DIC = '/home/sci/blakez/Reflect Affine/DIC_to_Reflect.txt'
src_pt = conf_dir + 'M{0}/section_{1}/{2}/section_{1}_confocal_relation_with_sidelight.txt'.format(mkyNum, secNum, region)
tar_pt = side_dir + 'M{0}/section_{1}/section_{1}_sidelight_relation_with_confocal.txt'.format(mkyNum, secNum)
# SID = '/home/sci/blakez/Reflect Affine/sidelight_to_DIC.txt'
src_im = common.LoadITKImage(conf_dir + 'M{0}/section_{1}/{3}/Ch{2}/M{0}_{1}_LGN_RHS_Ch{2}_z00.tif'.format(mkyNum, secNum, channel, region))
# tar_im = common.LoadITKImage('M{0}/{1}/Crop_ThirdNerve_EGFP_z16.tiff'.format(mkyNum, secNum))
# The points need to be chosen in the origin corrected sidescape for downstream purposes
affine = load_and_solve(tar_pt, src_pt)
out_grid = bb_grid_solver(src_im, affine)
z_stack = []
num_slices = len(glob.glob(conf_dir + 'M{0}/section_{1}/{3}/Ch{2}/*'.format(mkyNum, secNum, channel, region)))
for z in range(0, num_slices):
src_im = common.LoadITKImage(conf_dir + 'M{0}/section_{1}/{4}/Ch{2}/M{0}_{1}_LGN_RHS_Ch{2}_z{3}.tif'.format(mkyNum, secNum, channel, str(z).zfill(2), region))
aff_im = ca.Image3D(out_grid, ca.MEM_HOST)
cc.ApplyAffineReal(aff_im, src_im, affine)
common.SaveITKImage(aff_im, save_dir + 'M{0}/section_{1}/{4}/Ch{2}/M{0}_01_section_{1}_LGN_RHS_Ch{2}_conf_aff_sidelight_z{3}.tiff'.format(mkyNum, secNum, channel, str(z).zfill(2), region))
z_stack.append(aff_im)
print('==> Done with {0}/{1}'.format(z, num_slices - 1))
stacked = cc.Imlist_to_Im(z_stack)
stacked.setSpacing(ca.Vec3Df(out_grid.spacing()[0], out_grid.spacing()[1], 0.03/num_slices))
common.SaveITKImage(stacked, save_dir + 'M{0}/section_{1}/{3}/Ch{2}/M{0}_01_section_{1}_Ch{2}_conf_aff_sidelight_stack.nrrd'.format(mkyNum, secNum, channel, region))
common.DebugHere()
if channel==0:
cc.WriteGrid(stacked.grid(), save_dir + 'M{0}/section_{1}/{2}/affine_registration_grid.txt'.format(mkyNum, secNum, region))
def DispImage(im, title=None,
sliceIdx=None, dim='z',
cmap='gray', newFig=True,
rng=None, t=False, log=False):
# if this volume is not a slice already, extract a slice
sz = im.size().tolist()
if sz[common.DIMMAP[dim]] > 1:
im = common.ExtractSliceIm(im, sliceIdx, dim)
im.toType(core.MEM_HOST)
# transfer to host memory if necessary
if im.memType() == core.MEM_DEVICE:
tmp = core.Image3D(im.grid(), core.MEM_HOST)
core.Copy(tmp, im)
im = tmp
# create the figure if requested
if newFig:
if title is None:
plt.figure()
else:
plt.figure(title)
plt.clf()
# set the range
if rng is None:
vmin = None
vmax = None
else:
vmin = rng[0]
vmax = rng[1]
if log:
norm = matplotlib.colors.LogNorm(vmin=vmin, vmax=vmax)
else:
norm = matplotlib.colors.Normalize(vmin=vmin, vmax=vmax)
# convert to numpy array
arr = np.squeeze(im.asnp().copy())
# transpose if requested
if t:
arr = arr.T
# display
plt.imshow(arr, cmap=cmap, vmin=vmin, vmax=vmax, norm=norm, interpolation='nearest')
plt.axis('tight')
plt.axis('image')
if title is not None:
plt.title(title)
plt.xticks([])
plt.yticks([])
plt.draw()
def Loader(cfOb, memT):
'''Function for loading all of the images. All images get normalized between 0 and 1'''
#Load Source Images
bfiSrc = cc.LoadColorMHA(
pth.expanduser(secOb.bfiSrcPath + cfOb.bfiSrcName), memT)
ssiSrc = common.LoadITKImage(
pth.expanduser(secOb.ssiSrcPath + cfOb.ssiSrcName), memT)
ssiSrc /= ca.Max(ssiSrc)
#Load Mask Image
bfiMsk = common.LoadITKImage(
pth.expanduser(secOb.bfiMskPath + cfOb.bfiMskName), memT)
bfiMsk /= ca.Max(bfiMsk)
bfiMsk.setGrid(bfiSrc.grid())
ssiMsk = common.LoadITKImage(
pth.expanduser(secOb.ssiMskPath + cfOb.ssiMskName), memT)
ssiMsk /= ca.Max(ssiMsk)
return ssiSrc, bfiSrc, ssiMsk, bfiMsk
def preprocess_image(image_pyca, histeq):
image_np = common.AsNPCopy(image_pyca)
nan_mask = np.isnan(image_np)
image_np[nan_mask] = 0
image_np /= np.amax(image_np)
# perform histogram equalization if needed
if histeq:
image_np[image_np != 0] = exposure.equalize_hist(image_np[image_np != 0])
return image_np
def HGMWriteOutput(cf, groupState, tDiscGroup, isReporter):
# save initial momenta for residual geodesics, p, for all individuals
for i in range(len(groupState.t)):
if tDiscGroup[i].J is not None:
common.SaveITKField(
tDiscGroup[i].p0, cf.io.outputPrefix +
str(tDiscGroup[i].subjectId).replace('.', '_') + "_p0.mhd")
# write individual's energy history
energyFilename = cf.io.outputPrefix + str(
tDiscGroup[i].subjectId).replace('.',
'_') + "ResidualEnergy.csv"
HGMWriteEnergyHistoryToFile(tDiscGroup[i].Energy, energyFilename)
# save initial image and momenta for group gedoesic
if isReporter:
common.SaveITKImage(groupState.I0, cf.io.outputPrefix + "I0.mhd")
common.SaveITKField(groupState.m0, cf.io.outputPrefix + "m0.mhd")
# write energy history
energyFilename = cf.io.outputPrefix + "TotalEnergyHistory.csv"
HGMWriteEnergyHistoryToFile(groupState.EnergyHistory, energyFilename)
def Load(spec, argv):
"""Load YAML file, validate it, given spec (inserting defaults) and convert
to attributes
"""
if len(argv) < 2:
print('# Usage: ' + argv[0] + ' <config.yaml>')
print('# Below is a sample config YAML file')
print(SpecToYAML(spec))
print('# PyCA Version: %s (%s)' %
(common.PyCAVersion(), common.PyCABranch()))
if '_resource' in spec:
# this won't be printed by default, but let's keep track
print("_resource: " + spec['_resource'])
raise MissingConfigError()
d = LoadYAMLDict(argv[1])
# TODO: process overrides (given on the cmdline as key=value pairs)
return MkConfig(d, spec)
def test_AddMasked(self, disp=False):
SetMem(self.hIm, 0.0)
SetMem(self.dIm, 0.0)
hIm2 = common.RandImage(self.sz,
nSig=1.0,
gSig=0.0,
mType=MEM_HOST,
sp=self.imSp)
dIm2 = hIm2.copy()
dIm2.toType(MEM_DEVICE)
Add(self.hIm, self.hRandIm, hIm2, self.hRandMask)
Add(self.dIm, self.dRandIm, dIm2, self.dRandMask)
self.TestIm(self.hIm, self.dIm, name='AddMasked', disp=disp)
def JacDetPlot(vf, title='Jac. Det.',
jd_max=10.0, cmap='PRGn',
nonpos_clr=(1.0, 0.0, 0.0, 1.0),
sliceIdx=None, dim='z',
isVF=True,
newFig=False):
"""
Plot the jacobian determinant using logmapped colors, and setting
zero or negative values to 'nonpos_clr'. jd_max is the maximum
value to display without clamping, and also defines the min value
as 1.0/jd_max to assure 1.0 is centered in the colormap. If vf is
a vector field, compute the jacobian determinant. If it is an
Image3D, assume it is the jacobian determinant to be plotted.
"""
if common.IsField3D(vf):
grid = vf.grid()
mType = vf.memType()
h = core.ManagedField3D(grid, mType)
jacdet = core.ManagedImage3D(grid, mType)
core.Copy(h, vf)
if isVF:
core.VtoH_I(h)
core.JacDetH(jacdet, h)
elif common.IsImage3D(vf):
jacdet = vf
else:
raise Exception('unknown input type to JacDetPlot, %s'%\
str(type(vf)))
jd_cmap = JacDetCMap(jd_max=jd_max, cmap=cmap,
nonpos_clr=nonpos_clr)
DispImage(jacdet, title=title,
sliceIdx=sliceIdx, dim=dim,
rng=[1.0/jd_max, jd_max],
cmap=jd_cmap,
log=True,
newFig=newFig)
def MatchingImageMomentaWriteOuput(cf, geodesicState, EnergyHistory, m0, n1):
grid = geodesicState.J0.grid()
mType = geodesicState.J0.memType()
# save momenta for the gedoesic
common.SaveITKField(geodesicState.p0, cf.io.outputPrefix + "p0.mhd")
# save matched momenta for the geodesic
if cf.vectormomentum.matchImOnly:
m0 = common.LoadITKField(cf.study.m, mType)
ca.CoAd(geodesicState.p, geodesicState.rhoinv, m0)
common.SaveITKField(geodesicState.p, cf.io.outputPrefix + "m1.mhd")
# momenta match energy
if cf.vectormomentum.matchImOnly:
vecdiff = ca.ManagedField3D(grid, mType)
ca.Sub_I(geodesicState.p, n1)
ca.Copy(vecdiff, geodesicState.p)
geodesicState.diffOp.applyInverseOperator(geodesicState.p)
momentaMatchEnergy = ca.Dot(vecdiff, geodesicState.p) / (
float(geodesicState.p0.nVox()) * geodesicState.SigmaSlope *
geodesicState.SigmaSlope)
# save energy
energyFilename = cf.io.outputPrefix + "testMomentaMatchEnergy.csv"
with open(energyFilename, 'w') as f:
print >> f, momentaMatchEnergy
# save matched image for the geodesic
tempim = ca.ManagedImage3D(grid, mType)
ca.ApplyH(tempim, geodesicState.J0, geodesicState.rhoinv)
common.SaveITKImage(tempim, cf.io.outputPrefix + "I1.mhd")
# save energy
energyFilename = cf.io.outputPrefix + "energy.csv"
MatchingImageMomentaWriteEnergyHistoryToFile(EnergyHistory, energyFilename)
def SaveSlice(fname, im, sliceIdx=None, dim='z',
cmap='gray', rng=None, t=True):
sliceIm = common.ExtractSliceIm(im, sliceIdx, dim)
sliceIm.toType(core.MEM_HOST)
if rng is None:
vmin = None
vmax = None
else:
vmin = rng[0]
vmax = rng[1]
if t:
plt.imsave(fname, np.squeeze(sliceIm.asnp()).T,
cmap=cmap, vmin=vmin, vmax=vmax)
else:
plt.imsave(fname, np.squeeze(sliceIm.asnp()),
cmap=cmap, vmin=vmin, vmax=vmax)
def __init__(self, methodName='runTest'):
super(CpuGpuTestCase, self).__init__(methodName)
self.cudaEnabled = (GetNumberOfCUDADevices() > 0)
if self.cudaEnabled:
# allowable average abs. diff
self.AvEps = 1e-6
# allowable max abs. diff
self.MaxEps = 1e-4
# image size
self.sz = np.array([127, 119])
# spacing
self.sp = np.array([1.5, 2.1])
# fluid parameters
self.fluidParams = [1.0, 1.0, 0.0]
self.vsz = np.append(self.sz, 2)
self.imSz = Vec3Di(int(self.sz[0]), int(self.sz[1]), 1)
self.imSp = Vec3Df(float(self.sp[0]), float(self.sp[1]), 1.0)
# set up grid
self.grid = GridInfo(self.imSz, self.imSp)
# set up host / device images
self.I0Arr = common.DrawEllipse(self.sz, self.sz / 2,
self.sz[0] / 4, self.sz[1] / 3)
self.I1Arr = common.DrawEllipse(self.sz, self.sz / 2,
self.sz[0] / 3, self.sz[1] / 4)
self.I0Arr = common.GaussianBlur(self.I0Arr, 1.5)
self.I1Arr = common.GaussianBlur(self.I1Arr, 1.5)
self.hI0Orig = common.ImFromNPArr(self.I0Arr,
mType=MEM_HOST,
sp=self.imSp)
self.hI1Orig = common.ImFromNPArr(self.I1Arr,
mType=MEM_HOST,
sp=self.imSp)
self.dI0Orig = common.ImFromNPArr(self.I0Arr,
mType=MEM_DEVICE,
sp=self.imSp)
self.dI1Orig = common.ImFromNPArr(self.I1Arr,
mType=MEM_DEVICE,
sp=self.imSp)
def DefSeriesIter(I, V, t, func, args):
grid = I.grid()
mType = I.memType()
tlen = len(t)
h = core.Field3D(grid, mType)
IDef = core.Image3D(grid, mType)
core.SetToIdentity(h)
scratchV1 = core.Field3D(grid, mType)
scratchV2 = core.Field3D(grid, mType)
rtnarr = []
for tidx in range(tlen):
curt = t[tidx]
h = common.ComposeDef(V, curt, inverse=True,
asVField=False,
scratchV1=scratchV1,
scratchV2=scratchV2)
core.ApplyH(IDef, I, h)
r = func(IDef, curt, *args)
rtnarr.append(r)
return rtnarr
def Matching(cf):
if cf.compute.useCUDA and cf.compute.gpuID is not None:
ca.SetCUDADevice(cf.compute.gpuID)
# prepare output directory
common.Mkdir_p(os.path.dirname(cf.io.outputPrefix))
# Output loaded config
if cf.io.outputPrefix is not None:
cfstr = Config.ConfigToYAML(MatchingConfigSpec, cf)
with open(cf.io.outputPrefix + "parsedconfig.yaml", "w") as f:
f.write(cfstr)
mType = ca.MEM_DEVICE if cf.compute.useCUDA else ca.MEM_HOST
I0 = common.LoadITKImage(cf.study.I0, mType)
I1 = common.LoadITKImage(cf.study.I1, mType)
#ca.DivC_I(I0,255.0)
#ca.DivC_I(I1,255.0)
grid = I0.grid()
ca.ThreadMemoryManager.init(grid, mType, 1)
#common.DebugHere()
# TODO: need to work on these
t = [x*1./cf.optim.nTimeSteps for x in range(cf.optim.nTimeSteps+1)]
checkpointinds = range(1,len(t))
checkpointstates = [(ca.Field3D(grid,mType),ca.Field3D(grid,mType)) for idx in checkpointinds]
p = MatchingVariables(I0,I1, cf.vectormomentum.sigma, t,checkpointinds, checkpointstates, cf.vectormomentum.diffOpParams[0], cf.vectormomentum.diffOpParams[1], cf.vectormomentum.diffOpParams[2], cf.optim.Niter, cf.optim.stepSize, cf.optim.maxPert, cf.optim.nTimeSteps, integMethod = cf.optim.integMethod, optMethod=cf.optim.method, nInv=cf.optim.NIterForInverse,plotEvery=cf.io.plotEvery, plotSlice = cf.io.plotSlice, quiverEvery = cf.io.quiverEvery, outputPrefix = cf.io.outputPrefix)
RunMatching(p)
# write output
if cf.io.outputPrefix is not None:
# reset all variables by shooting once, may have been overwritten
CAvmCommon.IntegrateGeodesic(p.m0,p.t,p.diffOp,\
p.m, p.g, p.ginv,\
p.scratchV1, p.scratchV2,p. scratchV3,\
p.checkpointstates, p.checkpointinds,\
Ninv=p.nInv, integMethod = p.integMethod)
common.SaveITKField(p.m0, cf.io.outputPrefix+"m0.mhd")
common.SaveITKField(p.ginv, cf.io.outputPrefix+"phiinv.mhd")
common.SaveITKField(p.g, cf.io.outputPrefix+"phi.mhd")
BFIfname = 'block{0}_reg_fillblanks_{1}_hd4.mha'.format(block, color)
# BFI = cc.LoadMHA(BFIdir + BFIfname, mType)
# outimage = common.ExtractSliceIm(BFI,100)
cd.Disp3Pane(BFI_aff)
if color in ['bw', 've', 'weight']:
BFIdef = ca.ManagedImage3D(MRIgrid, mType) # these should be small enough
else:
BFIdef = ca.ManagedField3D(MRIgrid, mType)
cc.ApplyHReal(BFIdef, BFI_aff, h)
if sz == MRIsizes[-1] and color == colors[-1]:
cd.Disp3Pane(BFIdef)
# write data
if Write:
if color == 'rgb':
fname = 'block{0}_as_MRI_rgba_{1}.mha'.format(block, sz)
cc.WriteColorMHA(BFIdef, outdir + fname)
fname = 'block{0}_as_MRI_rgb_{1}.mha'.format(block, sz)
cc.WriteMHA(BFIdef, outdir + fname)
else:
#fname = 'block{0}_as_MRI_{1}_{2}_NEWLANDMARKS.mha'.format(block, color, sz)
fname = 'M15_01_to_MRI_TPS_bw_256_VE.mha'
cc.WriteMHA(BFIdef, outdir + fname)
cc.WriteMHA(h, outdir + 'M15_01_to_MRI_TPS_def_256.mha')
# cc.WriteMHA(h, outdir + 'block{0}_TPS_HField_{1}.mha'.format(block,sz))
cd.Disp3Pane(BFIdef)
common.DebugHere()
del BFIdef, BFI
del h
def main():
secNum = sys.argv[1]
mkyNum = sys.argv[2]
region = str(sys.argv[3])
# channel = sys.argv[3]
ext = 'M{0}/section_{1}/{2}/'.format(mkyNum, secNum, region)
ss_dir = '/home/sci/blakez/korenbergNAS/3D_database/Working/Microscopic/side_light_microscope/'
conf_dir = '/home/sci/blakez/korenbergNAS/3D_database/Working/Microscopic/confocal/'
memT = ca.MEM_DEVICE
try:
with open(
ss_dir +
'src_registration/M{0}/section_{1}/M{0}_01_section_{1}_regions.txt'
.format(mkyNum, secNum), 'r') as f:
region_dict = json.load(f)
f.close()
except IOError:
region_dict = {}
region_dict[region] = {}
region_dict['size'] = map(
int,
raw_input("What is the size of the full resolution image x,y? ").
split(','))
region_dict[region]['bbx'] = map(
int,
raw_input(
"What are the x indicies of the bounding box (Matlab Format x_start,x_stop? "
).split(','))
region_dict[region]['bby'] = map(
int,
raw_input(
"What are the y indicies of the bounding box (Matlab Format y_start,y_stop? "
).split(','))
if region not in region_dict:
region_dict[region] = {}
region_dict[region]['bbx'] = map(
int,
raw_input(
"What are the x indicies of the bounding box (Matlab Format x_start,x_stop? "
).split(','))
region_dict[region]['bby'] = map(
int,
raw_input(
"What are the y indicies of the bounding box (Matlab Format y_start,y_stop? "
).split(','))
img_region = common.LoadITKImage(
ss_dir +
'src_registration/M{0}/section_{1}/M{0}_01_section_{1}_{2}.tiff'.
format(mkyNum, secNum, region), ca.MEM_HOST)
ssiSrc = common.LoadITKImage(
ss_dir +
'src_registration/M{0}/section_{1}/frag0/M{0}_01_ssi_section_{1}_frag0.nrrd'
.format(mkyNum, secNum), ca.MEM_HOST)
bfi_df = common.LoadITKField(
ss_dir +
'Blockface_registered/M{0}/section_{1}/frag0/M{0}_01_ssi_section_{1}_frag0_to_bfi_real.mha'
.format(mkyNum, secNum), ca.MEM_DEVICE)
# Figure out the same region in the low resolution image: There is a transpose from here to matlab so dimensions are flipped
low_sz = ssiSrc.size().tolist()
yrng_raw = [(low_sz[1] * region_dict[region]['bbx'][0]) /
np.float(region_dict['size'][0]),
(low_sz[1] * region_dict[region]['bbx'][1]) /
np.float(region_dict['size'][0])]
xrng_raw = [(low_sz[0] * region_dict[region]['bby'][0]) /
np.float(region_dict['size'][1]),
(low_sz[0] * region_dict[region]['bby'][1]) /
np.float(region_dict['size'][1])]
yrng = [np.int(np.floor(yrng_raw[0])), np.int(np.ceil(yrng_raw[1]))]
xrng = [np.int(np.floor(xrng_raw[0])), np.int(np.ceil(xrng_raw[1]))]
low_sub = cc.SubVol(ssiSrc, xrng, yrng)
# Figure out the grid for the sub region in relation to the sidescape
originout = [
ssiSrc.origin().x + ssiSrc.spacing().x * xrng[0],
ssiSrc.origin().y + ssiSrc.spacing().y * yrng[0], 0
]
spacingout = [
(low_sub.size().x * ssiSrc.spacing().x) / (img_region.size().x),
(low_sub.size().y * ssiSrc.spacing().y) / (img_region.size().y), 1
]
gridout = cc.MakeGrid(img_region.size().tolist(), spacingout, originout)
img_region.setGrid(gridout)
only_sub = np.zeros(ssiSrc.size().tolist()[0:2])
only_sub[xrng[0]:xrng[1], yrng[0]:yrng[1]] = np.squeeze(low_sub.asnp())
only_sub = common.ImFromNPArr(only_sub)
only_sub.setGrid(ssiSrc.grid())
# Deform the only sub region to
only_sub.toType(ca.MEM_DEVICE)
def_sub = ca.Image3D(bfi_df.grid(), bfi_df.memType())
cc.ApplyHReal(def_sub, only_sub, bfi_df)
def_sub.toType(ca.MEM_HOST)
# Now have to find the bounding box in the deformation space (bfi space)
if 'deformation_bbx' not in region_dict[region]:
bb_def = np.squeeze(pp.LandmarkPicker([np.squeeze(def_sub.asnp())]))
bb_def_y = [bb_def[0][0], bb_def[1][0]]
bb_def_x = [bb_def[0][1], bb_def[1][1]]
region_dict[region]['deformation_bbx'] = bb_def_x
region_dict[region]['deformation_bby'] = bb_def_y
with open(
ss_dir +
'src_registration/M{0}/section_{1}/M{0}_01_section_{1}_regions.txt'
.format(mkyNum, secNum), 'w') as f:
json.dump(region_dict, f)
f.close()
# Now need to extract the region and create a deformation and image that have the same resolution as the img_region
deform_sub = cc.SubVol(bfi_df, region_dict[region]['deformation_bbx'],
region_dict[region]['deformation_bby'])
common.DebugHere()
sizeout = [
int(
np.ceil((deform_sub.size().x * deform_sub.spacing().x) /
img_region.spacing().x)),
int(
np.ceil((deform_sub.size().y * deform_sub.spacing().y) /
img_region.spacing().y)), 1
]
region_grid = cc.MakeGrid(sizeout,
img_region.spacing().tolist(),
deform_sub.origin().tolist())
def_im_region = ca.Image3D(region_grid, deform_sub.memType())
up_deformation = ca.Field3D(region_grid, deform_sub.memType())
img_region.toType(ca.MEM_DEVICE)
cc.ResampleWorld(up_deformation, deform_sub,
ca.BACKGROUND_STRATEGY_PARTIAL_ZERO)
cc.ApplyHReal(def_im_region, img_region, up_deformation)
ss_out = ss_dir + 'Blockface_registered/M{0}/section_{1}/{2}/'.format(
mkyNum, secNum, region)
if not pth.exists(pth.expanduser(ss_out)):
os.mkdir(pth.expanduser(ss_out))
common.SaveITKImage(
def_im_region,
pth.expanduser(ss_out) +
'M{0}_01_section_{1}_{2}_def_to_bfi.nrrd'.format(
mkyNum, secNum, region))
common.SaveITKImage(
def_im_region,
pth.expanduser(ss_out) +
'M{0}_01_section_{1}_{2}_def_to_bfi.tiff'.format(
mkyNum, secNum, region))
del img_region, def_im_region, ssiSrc, deform_sub
# Now apply the same deformation to the confocal images
conf_grid = cc.LoadGrid(
conf_dir +
'sidelight_registered/M{0}/section_{1}/{2}/affine_registration_grid.txt'
.format(mkyNum, secNum, region))
cf_out = conf_dir + 'blockface_registered/M{0}/section_{1}/{2}/'.format(
mkyNum, secNum, region)
# confocal.toType(ca.MEM_DEVICE)
# def_conf = ca.Image3D(region_grid, deform_sub.memType())
# cc.ApplyHReal(def_conf, confocal, up_deformation)
for channel in range(0, 4):
z_stack = []
num_slices = len(
glob.glob(conf_dir +
'sidelight_registered/M{0}/section_{1}/{3}/Ch{2}/*.tiff'.
format(mkyNum, secNum, channel, region)))
for z in range(0, num_slices):
src_im = common.LoadITKImage(
conf_dir +
'sidelight_registered/M{0}/section_{1}/{3}/Ch{2}/M{0}_01_section_{1}_LGN_RHS_Ch{2}_conf_aff_sidelight_z{4}.tiff'
.format(mkyNum, secNum, channel, region,
str(z).zfill(2)))
src_im.setGrid(
cc.MakeGrid(
ca.Vec3Di(conf_grid.size().x,
conf_grid.size().y, 1), conf_grid.spacing(),
conf_grid.origin()))
src_im.toType(ca.MEM_DEVICE)
def_im = ca.Image3D(region_grid, ca.MEM_DEVICE)
cc.ApplyHReal(def_im, src_im, up_deformation)
def_im.toType(ca.MEM_HOST)
common.SaveITKImage(
def_im, cf_out +
'Ch{2}/M{0}_01_section_{1}_{3}_Ch{2}_conf_def_blockface_z{4}.tiff'
.format(mkyNum, secNum, channel, region,
str(z).zfill(2)))
if z == 0:
common.SaveITKImage(
def_im, cf_out +
'Ch{2}/M{0}_01_section_{1}_{3}_Ch{2}_conf_def_blockface_z{4}.nrrd'
.format(mkyNum, secNum, channel, region,
str(z).zfill(2)))
z_stack.append(def_im)
print('==> Done with Ch {0}: {1}/{2}'.format(
channel, z, num_slices - 1))
stacked = cc.Imlist_to_Im(z_stack)
stacked.setSpacing(
ca.Vec3Df(region_grid.spacing().x,
region_grid.spacing().y,
conf_grid.spacing().z))
common.SaveITKImage(
stacked, cf_out +
'Ch{2}/M{0}_01_section_{1}_{3}_Ch{2}_conf_def_blockface_stack.nrrd'
.format(mkyNum, secNum, channel, region))
if channel == 0:
cc.WriteGrid(
stacked.grid(),
cf_out + 'deformed_registration_grid.txt'.format(
mkyNum, secNum, region))
