def get_shifts(md, data_shape, x, y):
x = round(x)
y = round(y)
# pixel size in nm
vx, vy, _ = md.voxelsize_nm
# position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(md)
x_ = (x + roi_x0) * vx
y_ = (y + roi_y0) * vy
# look up shifts
if not md.getOrDefault('Analysis.FitShifts', False):
DeltaX = md.chroma.dx.ev(x_, y_)
DeltaY = md.chroma.dy.ev(x_, y_)
else:
DeltaX = 0
DeltaY = 0
# find shift in whole pixels
dxp = int(DeltaX / vx)
dyp = int(DeltaY / vy)
return DeltaX, DeltaY, dxp, dxy
def OnGenShiftmapQuad(self, event):
from PYME.Analysis.points import twoColour, twoColourPlot
from PYME.IO.MetaDataHandler import get_camera_roi_origin
pipeline = self.visFr.pipeline
vs = pipeline.mdh.voxelsize_nm
roi_x0, roi_y0 = get_camera_roi_origin(pipeline.mdh)
x0 = (roi_x0)*vs[0]
y0 = (roi_y0)*vs[1]
lx = len(pipeline.filter['x'])
bbox = None#[0,(pipeline.mdh['Camera.ROIWidth'] + 1)*vs[0], 0,(pipeline.mdh['Camera.ROIHeight'] + 1)*vs[1]]
dx, dy, spx, spy, good = twoColour.genShiftVectorFieldQ(pipeline.filter['x']+.1*np.random.randn(lx) + x0, pipeline.filter['y']+.1*np.random.randn(lx) + y0, pipeline.filter['fitResults_dx'], pipeline.filter['fitResults_dy'], pipeline.filter['fitError_dx'], pipeline.filter['fitError_dy'], bbox=bbox)
#twoColourPlot.PlotShiftField(dx, dy, spx, spy)
twoColourPlot.PlotShiftField2(spx, spy, pipeline.mdh['Splitter.Channel0ROI'][2:], voxelsize=vs)
twoColourPlot.PlotShiftResiduals(pipeline['x'][good] + x0, pipeline['y'][good] + y0, pipeline['fitResults_dx'][good], pipeline['fitResults_dy'][good], spx, spy)
from six.moves import cPickle
defFile = os.path.splitext(os.path.split(self.visFr.GetTitle())[-1])[0] + '.sf'
fdialog = wx.FileDialog(None, 'Save shift field as ...',
wildcard='Shift Field file (*.sf)|*.sf', style=wx.FD_SAVE, defaultDir = nameUtils.genShiftFieldDirectoryPath(), defaultFile=defFile)
succ = fdialog.ShowModal()
if (succ == wx.ID_OK):
fpath = fdialog.GetPath()
#save as a pickle containing the data and voxelsize
fid = open(fpath, 'wb')
cPickle.dump((spx, spy), fid, 2)
fid.close()
def remap_splitter_coords(md, data_shape, x, y):
vx, vy, _ = md.voxelsize_nm
x0, y0 = get_camera_roi_origin(md)
if False: #'Splitter.Channel0ROI' in md.getEntryNames():
xg, yg, w, h = md['Splitter.Channel0ROI']
xr, yr, w, h = md['Splitter.Channel1ROI']
else:
xg, yg, w, h = 0, 0, data_shape[0], data_shape[1]
xr, yr = w, h
xn = x + (xr - xg)
yn = y + (yr - yg)
if md.get(
'Splitter.Flip', True
): #not (('Splitter.Flip' in md.getEntryNames() and not md.getEntry('Splitter.Flip'))):
yn = (h - y0 - y) + yr - yg
#chromatic shift
if 'chroma.dx' in md.getEntryNames():
dx = md['chroma.dx'].ev((x + x0) * vx, (y + y0) * vy) / vx
dy = md['chroma.dy'].ev((x + x0) * vx, (y + y0) * vy) / vy
xn -= dx
yn -= dy
return xn, yn
def _parseROI(md):
"""
Extract ROI coordinates from metadata
TODO - refactor out of here as it is being used in non-fitting code
Parameters
----------
md: dict-like
Metadata containing Camera.ROIWidth, Camera.ROIHeight or multiview information
Returns
-------
roi_slices: list
list of (x, y) slices to extract ROI from camera maps
"""
if any(md.get('Multiview.ActiveViews', [])):
#special case handling for multiview ROIs
origins = [
md['Multiview.ROI%dOrigin' % ind]
for ind in md['Multiview.ActiveViews']
]
size_x, size_y = md['Multiview.ROISize']
return [(slice(int(ox),
int(ox + size_x)), slice(int(oy), int(oy + size_y)))
for ox, oy in origins]
x0, y0 = get_camera_roi_origin(md)
x1 = x0 + md['Camera.ROIWidth']
y1 = y0 + md['Camera.ROIHeight']
return [(slice(int(x0), int(x1)), slice(int(y0), int(y1)))]
def __getitem__(self, key):
#print key
xslice, yslice, zslice = key
#cut region out of data stack
dataROI = self.data[xslice, yslice, zslice]
#average in z
#dataMean = dataROI.mean(2) - self.metadata.CCD.ADOffset
#generate grid to evaluate function on
Xg, Yg = scipy.mgrid[xslice, yslice]
vs = self.metadata.voxelsize_nm
Xg = vs.x*Xg
Yg = vs.y*Yg
#generate a corrected grid for the red channel
#note that we're cheating a little here - for shifts which are slowly
#varying we should be able to set Xr = Xg + delta_x(\bar{Xr}) and
#similarly for y. For slowly varying shifts the following should be
#equivalent to this. For rapidly varying shifts all bets are off ...
#DeltaX, DeltaY = twoColour.getCorrection(Xg.mean(), Yg.mean(), self.metadata.chroma.dx,self.metadata.chroma.dy)
roi_x0, roi_y0 = get_camera_roi_origin(self.metadata)
x_ = Xg.mean() + roi_x0*vs.x
y_ = Yg.mean() + roi_y0*vs.y
DeltaX = self.metadata.chroma.dx.ev(x_, y_)
DeltaY = self.metadata.chroma.dy.ev(x_, y_)
Xr = Xg + DeltaX
Yr = Yg + DeltaY
if not self.background is None and len(numpy.shape(self.background)) > 1 and not ('Analysis.subtractBackground' in self.metadata.getEntryNames() and self.metadata.Analysis.subtractBackground == False):
bgROI = self.background[xslice, yslice, zslice]
dataROI = dataROI - bgROI
Ag = dataROI[:,:,0]
Ar = dataROI[:,:,1]
#print Xg.shape, Ag.shape
x0 = (Xg*Ag + Xr*Ar).sum()/(Ag.sum() + Ar.sum())
y0 = (Yg*Ag + Yr*Ar).sum()/(Ag.sum() + Ar.sum())
sig_xl = (numpy.maximum(0, x0 - Xg)*Ag + numpy.maximum(0, x0 - Xr)*Ar).sum()/(Ag.sum() + Ar.sum())
sig_xr = (numpy.maximum(0, Xg - x0)*Ag + numpy.maximum(0, Xr - x0)*Ar).sum()/(Ag.sum() + Ar.sum())
sig_yu = (numpy.maximum(0, y0 - Yg)*Ag + numpy.maximum(0, y0 - Yr)*Ar).sum()/(Ag.sum() + Ar.sum())
sig_yd = (numpy.maximum(0, Yg - y0)*Ag + numpy.maximum(0, Yr - y0)*Ar).sum()/(Ag.sum() + Ar.sum())
Ag = Ag.sum() #amplitude
Ar = Ar.sum() #amplitude
res = numpy.array([Ag, Ar, x0, y0, sig_xl, sig_xr, sig_yu, sig_yd])
return COIFitResultR(res, self.metadata, (xslice, yslice, zslice))
def OnApplyShiftmap(self, event):
"""apply a vectorial correction for chromatic shift to an image - this
is a generic vectorial shift compensation, rather than the secial case
correction used with the splitter."""
from scipy import ndimage
import numpy as np
from PYME.DSView import ImageStack, ViewIm3D
from PYME.IO.MetaDataHandler import get_camera_roi_origin
dlg = ShiftmapSelectionDialog(self.dsviewer, self.image)
succ = dlg.ShowModal()
if (succ == wx.ID_OK):
#self.ds = example.CDataStack(fdialog.GetPath().encode())
#self.ds =
ds = []
shiftFiles = {}
X, Y, Z = np.mgrid[0:self.image.data.shape[0], 0:self.image.data.shape[1], 0:self.image.data.shape[2]]
vx, vy, vz = self.image.voxelsize_nm
roi_x0, roi_y0 = get_camera_roi_origin(self.image.mdh)
for ch in range(self.image.data.shape[3]):
sfFilename = dlg.GetChanFilename(ch)
shiftFiles[ch] = sfFilename
data = self.image.data[:,:,:, ch]
if os.path.exists(sfFilename):
spx, spy, dz = np.load(sfFilename)
dx = spx.ev(vx*(X+roi_x0), vy*(Y+roi_y0))/vx
dy = spy.ev(vx*(X+roi_x0), vy*(Y+roi_y0))/vy
dz = dz/vz
ds.append(ndimage.map_coordinates(data, [X+dx, Y+dy, Z+dz], mode='nearest'))
else:
ds.append(data)
fns = os.path.split(self.image.filename)[1]
im = ImageStack(ds, titleStub = '%s - corrected' % fns)
im.mdh.copyEntriesFrom(self.image.mdh)
im.mdh['Parent'] = self.image.filename
im.mdh.setEntry('ChromaCorrection.ShiftFilenames', shiftFiles)
if 'fitResults' in dir(self.image):
im.fitResults = self.image.fitResults
#im.mdh['Processing.GaussianFilter'] = sigmas
if self.dsviewer.mode == 'visGUI':
mode = 'visGUI'
else:
mode = 'lite'
dv = ViewIm3D(im, mode=mode, glCanvas=self.dsviewer.glCanvas, parent=wx.GetTopLevelParent(self.dsviewer))
dlg.Destroy()
def insert_into_full_map(dark, variance, metadata, sensor_size=(2048, 2048)):
"""
Embeds partial-sensor camera maps into full-sized camera map by padding with basic values in metadata. Alternatively
can be used to create boring maps to use in place of metadata scalars.
Parameters
----------
dark: ndarray or None
darkmap for valid ROI, or None to generate a uniform, ~useless metadata map
variance: ndarray
variance for valid ROI, or None to generate a uniform, ~useless metadata map
metadata: dict-like
ROI informatrion and camera noise parameters to use when padding maps
sensor_size: 2-int tuple
x and y camera sensor size
Returns
-------
full_dark: ndarray
padded dark map
full_var: ndarray
padded variance map
mdh: PYME.IO.MetadataHandler.NestedClassMDHandler
metadata handler to be associated with full maps while maintaining information about the original/valid ROI.
"""
mdh = NestedClassMDHandler()
mdh.copyEntriesFrom(metadata)
mdh.setEntry('Analysis.name', 'mean-variance')
x_origin, y_origin = get_camera_roi_origin(mdh)
mdh.setEntry('Analysis.valid.ROIOriginX', x_origin)
mdh.setEntry('Analysis.valid.ROIOriginY', y_origin)
mdh.setEntry('Analysis.valid.ROIWidth', mdh['Camera.ROIWidth'])
mdh.setEntry('Analysis.valid.ROIHeight', mdh['Camera.ROIHeight'])
mdh['Camera.ROIOriginX'], mdh['Camera.ROIOriginY'] = 0, 0
mdh['Camera.ROIWidth'], mdh['Camera.ROIHeight'] = sensor_size
mdh['Camera.ROI'] = (0, 0, sensor_size[0], sensor_size[1])
if dark is not None and variance is not None:
full_dark = mdh['Camera.ADOffset'] * np.ones(sensor_size,
dtype=dark.dtype)
full_var = (mdh['Camera.ReadNoise']**2) * np.ones(sensor_size,
dtype=variance.dtype)
xslice = slice(x_origin, x_origin + metadata['Camera.ROIWidth'])
yslice = slice(y_origin, y_origin + metadata['Camera.ROIHeight'])
full_dark[xslice, yslice] = dark
full_var[xslice, yslice] = variance
else:
logger.warning('Generating uniform maps')
full_dark = mdh['Camera.ADOffset'] * np.ones(sensor_size)
full_var = (mdh['Camera.ReadNoise']**2) * np.ones(sensor_size)
return full_dark, full_var, mdh
def __init__(self, parentSource, mdh, flatfield=None, dark=None):
self.source = parentSource
self.mdh = mdh
self.mdh['IntensityUnits'] = 'ADU'
x0, y0 = get_camera_roi_origin(mdh)
x1 = x0 + mdh.getOrDefault('Camera.ROIWidth', self.source.shape[0]) + 1
y1 = y0 + mdh.getOrDefault('Camera.ROIHeight',
self.source.shape[1]) + 1
self.flat = flatfield[x0:x1, y0:y1] if flatfield is not None else 1.
if dark is None:
self.dark = float(self.mdh.getEntry('Camera.ADOffset'))
else:
self.dark = dark[x0:x1, y0:y1].astype(float)
def get_labels_from_image(inp, img):
"""
Parameters
----------
inp: PYME.IO.tabular
localizations to query membership in labels from corresponding image
img: image.ImageStack
image containing labels to apply to localizations
Returns
-------
ids: ndarray
Label number from image, mapped to each localization within that label
numPerObject: ndarray
Number of localizations within the label that a given localization belongs to
"""
im_ox, im_oy, im_oz = img.origin
# account for ROIs
try:
p_ox, p_oy = np.array(get_camera_roi_origin(inp.mdh)) * np.array(
inp.mdh.voxelsize_nm[:2])
except AttributeError:
raise UserWarning('get_labels_from_image requires metadata')
pixX = np.round((inp['x'] + p_ox - im_ox) / img.pixelSize).astype('i')
pixY = np.round((inp['y'] + p_oy - im_oy) / img.pixelSize).astype('i')
pixZ = np.round((inp['z'] - im_oz) / img.sliceSize).astype('i')
if img.data.shape[2] == 1:
# disregard z for 2D images
pixZ = np.zeros_like(pixX)
ind = (pixX < img.data.shape[0]) * (pixY < img.data.shape[1]) * (
pixX >= 0) * (pixY >= 0) * (pixZ >= 0) * (pixZ < img.data.shape[2])
ids = np.zeros_like(pixX)
# assume there is only one channel
ids[ind] = np.atleast_3d(img.data[:, :, :,
0].squeeze())[pixX[ind], pixY[ind],
pixZ[ind]].astype('i')
numPerObject, b = np.histogram(ids, np.arange(ids.max() + 1.5) + .5)
return ids, numPerObject
def genFitImage(fitResults, metadata, fitfcn=f_Interp3d):
from PYME.IO.MetaDataHandler import get_camera_roi_origin
xslice = slice(*fitResults['slicesUsed']['x'])
yslice = slice(*fitResults['slicesUsed']['y'])
vx, vy = metadata.voxelsize_nm
#position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(metadata)
x_ = (xslice.start + roi_x0) * vx
y_ = (yslice.start + roi_y0) * vy
im = PSFFitFactory._evalModel(fitResults['fitResults'], metadata, xslice,
yslice, x_, y_)
return im[0].squeeze()
def genFitImage(fitResults, metadata):
xslice = slice(*fitResults['slicesUsed']['x'])
yslice = slice(*fitResults['slicesUsed']['y'])
vx, vy, _ = metadata.voxelsize_nm
#position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(metadata)
x_ = (xslice.start + roi_x0) * vx
y_ = (yslice.start + roi_y0) * vy
ratio = fitResults['ratio']
im = InterpFitFactory._evalModel(np.array(list(fitResults['fitResults'])), metadata, xslice, yslice, ratio, x_, y_)[0]
#print im.shape
return np.hstack([im[:,:,0], im[:,:,1]]).squeeze()
def map_splitter_coords(md, data_shape, x, y):
vx = md['voxelsize.x'] * 1e3
vy = md['voxelsize.y'] * 1e3
x0, y0 = get_camera_roi_origin(md)
if False: #'Splitter.Channel0ROI' in md.getEntryNames():
xg, yg, wg, hg = md['Splitter.Channel0ROI']
xr, yr, wr, hr = md['Splitter.Channel1ROI']
#w2 = w - x0
#h2 = h - y0
else:
xg, yg, wg, hg = 0, 0, data_shape[0], data_shape[1]
xr, yr, wr, hr = wg, hg, wg, hg
#xg, wg = _bdsClip(xg, wg, x0, data_shape[0])
#xr, wr = _bdsClip(xr, wr, x0, data_shape[0])
#yg, hg = _bdsClip(yg, hg, y0, data_shape[1])
#yr, hr = _bdsClip(yr, hr, y0, data_shape[1])
w = min(wg, wr)
h = min(hg, hr)
ch1 = (x >= xr) & (y >= yr)
xn = x - ch1 * xr
yn = y - ch1 * yr
if md.get(
'Splitter.Flip', True
): #not (('Splitter.Flip' in md.getEntryNames() and not md.getEntry('Splitter.Flip'))):
#yn = y - ch1*yr
yn += ch1 * (h - 2 * yn)
#chromatic shift
if 'chroma.dx' in md.getEntryNames():
dx = md['chroma.dx'].ev((xn + x0) * vx, (yn + y0) * vy) / vx
dy = md['chroma.dy'].ev((xn + x0) * vy, (yn + y0) * vy) / vy
xn += dx * ch1
yn += dy * ch1
return np.clip(xn, 0, w - 1), np.clip(yn, 0, h - 1)
def insertIntoFullMap(m, ve, smdh, chipsize=(2048, 2048)):
x0, y0 = get_camera_roi_origin(smdh)
validROI = {
'PosX': x0 + 1,
'PosY': x0 + 1,
'Width': smdh['Camera.ROIWidth'],
'Height': smdh['Camera.ROIHeight']
}
bmdh = NestedClassMDHandler()
bmdh.copyEntriesFrom(smdh)
bmdh.setEntry('Analysis.name', 'mean-variance')
bmdh.setEntry('Analysis.valid.ROIPosX', validROI['PosX'])
bmdh.setEntry('Analysis.valid.ROIPosY', validROI['PosY'])
bmdh.setEntry('Analysis.valid.ROIWidth', validROI['Width'])
bmdh.setEntry('Analysis.valid.ROIHeight', validROI['Height'])
bmdh['Camera.ROIOriginX'] = 0
bmdh['Camera.ROIOriginY'] = 0
bmdh['Camera.ROIWidth'] = chipsize[0]
bmdh['Camera.ROIHeight'] = chipsize[1]
bmdh['Camera.ROI'] = (0, 0, chipsize[0], chipsize[1])
if m is None:
mfull = np.zeros(chipsize, dtype='float64')
vefull = np.zeros(chipsize, dtype='float64')
else:
mfull = np.zeros(chipsize, dtype=m.dtype)
vefull = np.zeros(chipsize, dtype=ve.dtype)
mfull.fill(smdh['Camera.ADOffset'])
vefull.fill(smdh['Camera.ReadNoise']**2)
if m is not None:
mfull[validROI['PosX'] - 1:validROI['PosX'] - 1 + validROI['Width'],
validROI['PosY'] - 1:validROI['PosY'] - 1 +
validROI['Height']] = m
vefull[validROI['PosX'] - 1:validROI['PosX'] - 1 + validROI['Width'],
validROI['PosY'] - 1:validROI['PosY'] - 1 +
validROI['Height']] = ve
return mfull, vefull, bmdh
def genFitImage(fitResults, metadata):
from PYME.IO.MetaDataHandler import get_camera_roi_origin
xslice = slice(*fitResults['slicesUsed']['x'])
yslice = slice(*fitResults['slicesUsed']['y'])
vx = 1e3*metadata.voxelsize.x
vy = 1e3*metadata.voxelsize.y
#position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(metadata)
x_ = (xslice.start + roi_x0)*vx
y_ = (yslice.start + roi_y0)*vy
#ratio = fitResults['ratio']
im = InterpFitFactory._evalModel(np.array(list(fitResults['fitResults'])), metadata, xslice, yslice, x_, y_)[0]
#print im.shape
return np.hstack([im[:,:,0], im[:,:,1]]).squeeze()
def get_splitter_rois(md, data_shape):
x0, y0 = get_camera_roi_origin(md)
if 'Splitter.Channel0ROI' in md.getEntryNames():
xg, yg, wg, hg = md['Splitter.Channel0ROI']
xr, yr, wr, hr = md['Splitter.Channel1ROI']
#print 'Have splitter ROIs'
else:
xg = 0
yg = 0
wg = data_shape[0]
hg = data_shape[1] / 2
xr = 0
yr = hg
wr = data_shape[0]
hr = data_shape[1] / 2
#print yr, hr
xg, wg = _bdsClip(xg, wg, x0, data_shape[0])
xr, wr = _bdsClip(xr, wr, x0, data_shape[0])
yg, hg = _bdsClip(yg, hg, y0, data_shape[1])
yr, hr = _bdsClip(yr, hr, y0, data_shape[1])
w = min(wg, wr)
h = min(hg, hr)
#print yg, hg, yr, hr
if ('Splitter.Flip' in md.getEntryNames()
and not md.getEntry('Splitter.Flip')):
step = 1
return (slice(xg, xg + w,
1), slice(xr, xr + w,
1), slice(yg, yg + h,
1), slice(yr, yr + h, step))
else:
step = -1
return (slice(xg, xg + w, 1), slice(xr, xr + w, 1),
slice(yg + hg - h, yg + hg, 1), slice(yr + h, yr - 1, step))
def __init__(self, parentSource, mdh, flatfield, dark=None):
#self.h5Filename = getFullFilename(h5Filename)#convert relative path to full path
#self.h5File = tables.openFile(self.h5Filename)
self.source = parentSource
self.mdh = mdh
#self.flat = flatfield
x0, y0 = get_camera_roi_origin(mdh)
x1 = x0 + mdh.getOrDefault('Camera.ROIWidth', self.source.shape[0]) + 1
y1 = y0 + mdh.getOrDefault('Camera.ROIHeight',
self.source.shape[1]) + 1
#print((x0, x1, y0, y1))
#self.offset = mdh.getEntry()
self.flat = flatfield[x0:x1, y0:y1]
if dark is None:
self.dark = self.mdh.getEntry('Camera.ADOffset')
else:
self.dark = dark[x0:x1, y0:y1]
def OnExtract(self, event):
from PYME.DSView import View3D
from PYME.IO.MetaDataHandler import get_camera_roi_origin
#print 'extracting ...'
mdh = self.image.mdh
#dark = deTile.genDark(self.vp.do.ds, self.image.mdh)
dark = mdh.getEntry('Camera.ADOffset')
#split = False
frames = mdh.getEntry('Protocol.PrebleachFrames')
dt = self.image.data[:, :,
frames[0]:frames[1]].astype('f').mean(2) - dark
roi_x0, roi_y0 = get_camera_roi_origin(mdh)
ROIX1 = roi_x0 + 1
ROIY1 = roi_y0 + 1
ROIX2 = ROIX1 + mdh.getEntry('Camera.ROIWidth')
ROIY2 = ROIY1 + mdh.getEntry('Camera.ROIHeight')
if self.split:
from PYME.Acquire.Hardware import splitter
unmux = splitter.Unmixer(
[mdh.getEntry('chroma.dx'),
mdh.getEntry('chroma.dy')], 1e3 * mdh.getEntry('voxelsize.x'))
dt = unmux.Unmix(dt, self.mixmatrix, 0,
[ROIX1, ROIY1, ROIX2, ROIY2])
View3D(dt, 'Prebleach Image')
else:
View3D(dt, 'Prebleach Image')
def getSplitROIAtPoint(self, x, y, z=None, roiHalfSize=5, axialHalfSize=15):
"""Helper fcn to extract ROI from frame at given x,y, point from a multi-channel image.
Returns:
Xg - x coordinates of pixels in ROI in nm (channel 1)
Yg - y coordinates of pixels in ROI (chanel 1)
Xr - x coordinates of pixels in ROI in nm (channel 2)
Yr - y coordinates of pixels in ROI (chanel 2)
data - raw pixel data of ROI
background - extimated background for ROI
sigma - estimated error (std. dev) of pixel values
xslice - x slice into original data array used to get ROI (channel 1)
yslice - y slice into original data array (channel 1)
xslice2 - x slice into original data array used to get ROI (channel 2)
yslice2 - y slice into original data array (channel 2)
"""
x = round(x)
y = round(y)
roiHalfSize = int(roiHalfSize)
#pixel size in nm
vx, vy, _ = self.metadata.voxelsize_nm
#position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(self.metadata)
x_ = (x + roi_x0)*vx
y_ = (y + roi_y0)*vy
#look up shifts
if not self.metadata.getOrDefault('Analysis.FitShifts', False):
DeltaX = self.shift_x.ev(x_, y_)
DeltaY = self.shift_y.ev(x_, y_)
else:
DeltaX = 0
DeltaY = 0
#find shift in whole pixels
dxp = int(DeltaX/vx)
dyp = int(DeltaY/vy)
#find ROI which works in both channels
x01 = max(x - roiHalfSize, max(0, dxp))
x11 = min(max(x01, x + roiHalfSize + 1), self.data.shape[0] + min(0, dxp))
x02 = x01 - dxp
x12 = x11 - dxp
y01 = max(y - roiHalfSize, max(0, dyp))
y11 = min(max(y + roiHalfSize + 1, y01), self.data.shape[1] + min(0, dyp))
y02 = y01 - dyp
y12 = y11 - dyp
xslice = slice(int(x01), int(x11))
xslice2 = slice(int(x02), int(x12))
yslice = slice(int(y01), int(y11))
yslice2 = slice(int(y02), int(y12))
#cut region out of data stack
dataROI = np.copy(self.data[xslice, yslice, 0:2])
dataROI[:,:,1] = self.data[xslice2, yslice2, 1]
if self.noiseSigma is None:
sigma = self._calc_sigma(dataROI)
else:
sigma = self.noiseSigma[xslice, yslice, 0:2]
sigma[:,:,1] = self.noiseSigma[xslice2, yslice2, 1]
sigma = ndimage.maximum_filter(sigma, [3,3,0])
if self.metadata.getOrDefault('Analysis.subtractBackground', True) :
#print 'bgs'
if not self.background is None and len(np.shape(self.background)) > 1:
bgROI = self.background[xslice, yslice, 0:2]
bgROI[:,:,1] = self.background[xslice2, yslice2, 1]
else:
bgROI = np.zeros_like(dataROI) + (self.background if self.background else 0)
else:
bgROI = np.zeros_like(dataROI)
#generate grid to evaluate function on
Xg = vx*(np.mgrid[xslice] + self.roi_offset[0])
Yg = vy*(np.mgrid[yslice] + self.roi_offset[1])
#generate a corrected grid for the red channel
#note that we're cheating a little here - for shifts which are slowly
#varying we should be able to set Xr = Xg + delta_x(\bar{Xr}) and
#similarly for y. For slowly varying shifts the following should be
#equivalent to this. For rapidly varying shifts all bets are off ...
#DeltaX, DeltaY = twoColour.getCorrection(Xg.mean(), Yg.mean(), self.metadata['chroma.dx'],self.metadata['chroma.dy'])
Xr = Xg + DeltaX - vx*dxp
Yr = Yg + DeltaY - vy*dyp
return Xg, Yg, Xr, Yr, dataROI, bgROI, sigma, xslice, yslice, xslice2, yslice2
def tile(ds,
xm,
ym,
mdh,
split=True,
skipMoveFrames=True,
shiftfield=None,
mixmatrix=[[1., 0.], [0., 1.]],
correlate=False,
dark=None,
flat=None):
frameSizeX, frameSizeY, numFrames = ds.shape[:3]
if split:
frameSizeY /= 2
nchans = 2
unmux = splitter.Unmixer(shiftfield, mdh.voxelsize_nm.x)
else:
nchans = 1
#x & y positions of each frame
xps = xm(np.arange(numFrames))
yps = ym(np.arange(numFrames))
if mdh.getOrDefault('CameraOrientation.FlipX', False):
xps = -xps
if mdh.getOrDefault('CameraOrientation.FlipY', False):
yps = -yps
#give some room at the edges
bufSize = 0
if correlate:
bufSize = 300
#convert to pixels
xdp = (bufSize + ((xps - xps.min()) /
(mdh.getEntry('voxelsize.x'))).round()).astype('i')
ydp = (bufSize + ((yps - yps.min()) /
(mdh.getEntry('voxelsize.y'))).round()).astype('i')
#print (xps - xps.min()), mdh.getEntry('voxelsize.x')
#work out how big our tiled image is going to be
imageSizeX = np.ceil(xdp.max() + frameSizeX + bufSize)
imageSizeY = np.ceil(ydp.max() + frameSizeY + bufSize)
#print imageSizeX, imageSizeY
#allocate an empty array for the image
im = np.zeros([imageSizeX, imageSizeY, nchans])
# and to record occupancy (to normalise overlapping tiles)
occupancy = np.zeros([imageSizeX, imageSizeY, nchans])
#calculate a weighting matrix (to allow feathering at the edges - TODO)
weights = np.ones((frameSizeX, frameSizeY, nchans))
#weights[:, :10, :] = 0 #avoid splitter edge artefacts
#weights[:, -10:, :] = 0
#print weights[:20, :].shape
edgeRamp = min(100, int(.5 * ds.shape[0]))
weights[:edgeRamp, :, :] *= np.linspace(0, 1, edgeRamp)[:, None, None]
weights[-edgeRamp:, :, :] *= np.linspace(1, 0, edgeRamp)[:, None, None]
weights[:, :edgeRamp, :] *= np.linspace(0, 1, edgeRamp)[None, :, None]
weights[:, -edgeRamp:, :] *= np.linspace(1, 0, edgeRamp)[None, :, None]
roi_x0, roi_y0 = get_camera_roi_origin(mdh)
ROIX1 = roi_x0 + 1
ROIY1 = roi_y0 + 1
ROIX2 = ROIX1 + mdh.getEntry('Camera.ROIWidth')
ROIY2 = ROIY1 + mdh.getEntry('Camera.ROIHeight')
if dark is None:
offset = float(mdh.getEntry('Camera.ADOffset'))
else:
offset = 0.
# #get a sorted list of x and y values
# xvs = list(set(xdp))
# xvs.sort()
#
# yvs = list(set(ydp))
# yvs.sort()
for i in range(mdh.getEntry('Protocol.DataStartsAt'), numFrames):
if xdp[i - 1] == xdp[i] or not skipMoveFrames:
d = ds[:, :, i].astype('f')
if not dark is None:
d = d - dark
if not flat is None:
d = d * flat
if split:
d = np.concatenate(
unmux.Unmix(d, mixmatrix, offset,
[ROIX1, ROIY1, ROIX2, ROIY2]), 2)
#else:
#d = d.reshape(list(d.shape) + [1])
imr = (im[xdp[i]:(xdp[i] + frameSizeX),
ydp[i]:(ydp[i] + frameSizeY), :] /
occupancy[xdp[i]:(xdp[i] + frameSizeX),
ydp[i]:(ydp[i] + frameSizeY), :])
alreadyThere = (weights *
occupancy[xdp[i]:(xdp[i] + frameSizeX), ydp[i]:
(ydp[i] + frameSizeY), :]).sum(2) > 0
#d_ = d.sum(2)
if split:
r0 = imr[:, :, 0][alreadyThere].sum()
r1 = imr[:, :, 1][alreadyThere].sum()
if r0 == 0:
r0 = 1
else:
r0 = r0 / (d[:, :, 0][alreadyThere]).sum()
if r1 == 0:
r1 = 1
else:
r1 = r1 / (d[:, :, 1][alreadyThere]).sum()
rt = np.array([r0, r1])
imr = imr.sum(2)
else:
rt = imr[:, :, 0][alreadyThere].sum()
if rt == 0:
rt = 1
else:
rt = rt / (d[:, :, 0][alreadyThere]).sum()
rt = np.array([rt])
#print rt
if correlate:
if (alreadyThere.sum() > 50):
dx = 0
dy = 0
rois = findRectangularROIs(alreadyThere)
for r in rois:
x0, y0, x1, y1 = r
#print r
dx_, dy_ = calcCorrShift(
d.sum(2)[x0:x1, y0:y1], imr[x0:x1, y0:y1])
print(('d_', dx_, dy_))
dx += dx_
dy += dy_
dx = np.round(dx / len(rois))
dy = np.round(dy / len(rois))
print((dx, dy))
#dx, dy = (0,0)
else:
dx, dy = (0, 0)
im[(xdp[i] + dx):(xdp[i] + frameSizeX + dx),
(ydp[i] + dy):(ydp[i] + frameSizeY +
dy), :] += weights * d * rt[None, None, :]
occupancy[(xdp[i] + dx):(xdp[i] + frameSizeX + dx),
(ydp[i] + dy):(ydp[i] + frameSizeY +
dy), :] += weights
else:
#print weights.shape, rt.shape, d.shape
im[xdp[i]:(xdp[i] + frameSizeX),
ydp[i]:(ydp[i] +
frameSizeY), :] += weights * d #*rt[None, None, :]
occupancy[xdp[i]:(xdp[i] + frameSizeX),
ydp[i]:(ydp[i] + frameSizeY), :] += weights
ret = (im / occupancy).squeeze()
#print ret.shape, occupancy.shape
ret[occupancy.squeeze() == 0] = 0 #fix up /0s
return ret
def _extractROI_1(self, fri):
from PYME.IO.MetaDataHandler import get_camera_roi_origin
roi_x0, roi_y0 = get_camera_roi_origin(self.mdh)
if 'Splitter' in self.mdh['Analysis.FitModule']:
# is a splitter fit
if 'Splitter.Channel0ROI' in self.mdh.getEntryNames():
x0, y0, w, h = self.mdh['Splitter.Channel0ROI']
x0 -= roi_x0
y0 -= roi_y0
# g = self.data[x0:(x0+w), y0:(y0+h)]
x1, y1, w, h = self.mdh['Splitter.Channel1ROI']
x1 -= roi_x0
y1 -= roi_y0
# r = self.data[x0:(x0+w), y0:(y0+h)]
else:
x0, y0 = 0, 0
x1, y1 = 0, (self.mdh['Camera.ROIHeight'] + 1) / 2
h = y1
slux = fri['slicesUsed']['x']
sluy = fri['slicesUsed']['y']
slx = slice(slux[0], slux[1])
sly = slice(sluy[0], sluy[1])
# sx0 = slice(x0+ slux[0], x0+slux[1])
# sy0 = slice(y0+ sluy[0], y0+sluy[1])
if 'NR' in self.mdh['Analysis.FitModule']:
# for fits which take chromatic shift into account when selecting ROIs
# pixel size in nm
vx, vy, _ = self.mdh.voxelsize_nm
# position in nm from camera origin
x_ = ((slux[0] + slux[1]) / 2. + roi_x0) * vx
y_ = ((sluy[0] + sluy[1]) / 2. + roi_y0) * vy
# look up shifts
if not self.mdh.getOrDefault('Analysis.FitShifts', False):
DeltaX = self.mdh.chroma.dx.ev(x_, y_)
DeltaY = self.mdh.chroma.dy.ev(x_, y_)
else:
DeltaX = 0
DeltaY = 0
# find shift in whole pixels
dxp = int(DeltaX / vx)
dyp = int(DeltaY / vy)
print((DeltaX, DeltaY, dxp, dyp))
x1 -= dxp
y1 -= dyp
sx1 = slice(x1 - x0 + slux[0], x1 - x0 + slux[1])
if ('Splitter.Flip' in self.mdh.getEntryNames()
and not self.mdh.getEntry('Splitter.Flip')):
sy1 = slice(y1 - y0 + sluy[0], y1 - y0 + sluy[1])
else:
sy1 = slice(y1 + h + y0 - sluy[0], y1 + h + y0 - sluy[1],
-1) # FIXME
print((slx, sx1, sly, sy1))
print(h, y0, y1, sluy)
g = self.ds[slx, sly, int(fri['tIndex'])].squeeze()
r = self.ds[sx1, sy1, int(fri['tIndex'])].squeeze()
return np.hstack([g, r])
else:
return self.ds[slice(*fri['slicesUsed']['x']),
slice(*fri['slicesUsed']['y']),
int(fri['tIndex'])].squeeze()
def get_labels_from_image(label_image, points, minimum_localizations=1):
"""
Function to extract labels from a segmented image (2D or 3D) at given locations.
Parameters
----------
label_image: PYME.IO.image.ImageStack instance
an image containing object labels
points: tabular-like (PYME.IO.tabular, np.recarray, pandas DataFrame) containing 'x', 'y' & 'z' columns
locations at which to extract labels
Returns
-------
ids: Label number from image, mapped to each localization within that label
numPerObject: Number of localizations within the label that a given localization belongs to
"""
from PYME.IO.MetaDataHandler import get_camera_roi_origin
im_ox, im_oy, im_oz = label_image.origin
# account for ROIs
try:
roi_x0, roi_y0 = get_camera_roi_origin(points.mdh)
vs = points.mdh.voxelsize_nm
p_ox = roi_x0 * vs.x
p_oy = roi_y0 * vs.y
except AttributeError:
raise RuntimeError(
'label image requires metadata specifying ROI position and voxelsize'
)
# Image origin is referenced to top-left corner of pixelated image.
# FIXME - localisations are currently referenced to centre of raw pixels
pixX = np.floor(
(points['x'] + p_ox - im_ox) / label_image.pixelSize).astype('i')
pixY = np.floor(
(points['y'] + p_oy - im_oy) / label_image.pixelSize).astype('i')
pixZ = np.floor((points['z'] - im_oz) / label_image.sliceSize).astype('i')
label_data = label_image.data
if label_data.shape[2] == 1:
# disregard z for 2D images
pixZ = np.zeros_like(pixX)
ind = (pixX < label_data.shape[0]) * (pixY < label_data.shape[1]) * (
pixX >= 0) * (pixY >= 0) * (pixZ >= 0) * (pixZ < label_data.shape[2])
ids = np.zeros_like(pixX)
# assume there is only one channel
ids[ind] = np.atleast_3d(label_data[:, :, :,
0].squeeze())[pixX[ind], pixY[ind],
pixZ[ind]].astype('i')
# check if we keep all labels
if minimum_localizations > 1: # skip if we don't need this
labels, counts = np.unique(ids, return_counts=True)
labels, counts = labels[1:], counts[
1:] # ignore unlabeled points, or zero-label
for label, count in zip(labels, counts):
if count < minimum_localizations:
ids[ids == label] = 0
numPerObject, b = np.histogram(ids, np.arange(ids.max() + 1.5) + .5)
return ids, numPerObject
def _getUSDataSources(self):
mdh = self.image.mdh
if 'chroma.dx' in mdh.getEntryNames():
sf = (mdh['chroma.dx'], mdh['chroma.dy'])
elif global_shiftfield:
sf = global_shiftfield
else:
sf = None
flip = True
if 'Splitter.Flip' in mdh.getEntryNames() and not mdh['Splitter.Flip']:
flip = False
chanROIs = None
if 'Splitter.Channel0ROI' in mdh.getEntryNames():
chanROIs = [
mdh['Splitter.Channel0ROI'], mdh['Splitter.Channel1ROI']
]
if 'Multiview.NumROIs' in mdh.getEntryNames():
# we have more than 2 ROIs
numROIs = mdh['Multiview.NumROIs']
w, h = mdh['Multiview.ROISize']
#print self.image.data.shape, w, h, numROIs
flip = False
if self.image.data.shape[0] == numROIs * w:
#we are extracted as expected.
h_ = min(h, int(self.image.data.shape[1]))
chanROIs = []
for i in range(numROIs):
x0, y0 = (i * w, 0)
chanROIs.append((x0, y0, w, h_))
#TODO - Fix me to use proper coordinates
ROIX1, ROIY1 = (1, 1)
ROIX2, ROIY2 = (w * numROIs, h_)
else:
#raw data - do the extraction ourselves
raise RuntimeError(
"data has not been unsplit, we can't handle this at present"
)
chanROIs = []
for i in range(numROIs):
x0, y0 = mdh['Multiview.ROISize']
chanROIs.append((x0, y0, w, h))
else:
#default to old splitter code
from PYME.IO.MetaDataHandler import get_camera_roi_origin
roi_x0, roi_y0 = get_camera_roi_origin(mdh)
ROIX1 = roi_x0 + 1
ROIY1 = roi_y0 + 1
ROIX2 = ROIX1 + mdh.getEntry('Camera.ROIWidth')
ROIY2 = ROIY1 + mdh.getEntry('Camera.ROIHeight')
numROIs = 2
usds = [
UnsplitDataSource.DataSource(self.image.data,
[ROIX1, ROIY1, ROIX2, ROIY2],
i,
flip,
sf,
chanROIs=chanROIs,
voxelsize=self.image.voxelsize)
for i in range(numROIs)
]
return usds
def getMultiviewROIAtPoint(self, x, y, z=None, roiHalfSize=5, axialHalfSize=15):
"""Helper fcn to extract ROI from frame at given x,y, point from a multi-channel image.
WARNING: EXPERIMENTAL WORK IN PROGRESS!!! This will eventually replace getSplitROIAtPoint and generalise to higher
dimensional splitting (e.g. 4-quadrant systems such as the 4Pi-SMS) but is not useful in it's current form.
Returns:
Xg - x coordinates of pixels in ROI in nm (channel 1)
Yg - y coordinates of pixels in ROI (chanel 1)
Xr - x coordinates of pixels in ROI in nm (channel 2)
Yr - y coordinates of pixels in ROI (chanel 2)
data - raw pixel data of ROI
background - extimated background for ROI
sigma - estimated error (std. dev) of pixel values
xslice - x slice into original data array used to get ROI (channel 1)
yslice - y slice into original data array (channel 1)
xslice2 - x slice into original data array used to get ROI (channel 2)
yslice2 - y slice into original data array (channel 2)
"""
x = round(x)
y = round(y)
#pixel size in nm
vx, vy, _ = self.metadata.voxelsize_nm
#position in nm from camera origin
roi_x0, roi_y0 = get_camera_roi_origin(self.metadata)
x_ = (x + roi_x0) * vx
y_ = (y + roi_y0) * vy
#look up shifts
if not self.metadata.getOrDefault('Analysis.FitShifts', False):
DeltaX = self.shift_x.ev(x_, y_)
DeltaY = self.shift_y.ev(x_, y_)
else:
DeltaX = 0
DeltaY = 0
#find shift in whole pixels
dxp = int(DeltaX / vx)
dyp = int(DeltaY / vy)
#find ROI which works in both channels
#if dxp < 0:
x01 = max(x - roiHalfSize, max(0, dxp))
x11 = min(max(x01, x + roiHalfSize + 1), self.data.shape[0] + min(0, dxp))
x02 = x01 - dxp
x12 = x11 - dxp
y01 = max(y - roiHalfSize, max(0, dyp))
y11 = min(max(y + roiHalfSize + 1, y01), self.data.shape[1] + min(0, dyp))
y02 = y01 - dyp
y12 = y11 - dyp
xslice = slice(int(x01), int(x11))
xslice2 = slice(int(x02), int(x12))
yslice = slice(int(y01), int(y11))
yslice2 = slice(int(y02), int(y12))
#print xslice2, yslice2
#cut region out of data stack
dataROI = self.data[xslice, yslice, 0:2]
#print dataROI.shape
dataROI[:, :, 1] = self.data[xslice2, yslice2, 1]
nSlices = 1
#sigma = np.sqrt(self.metadata.Camera.ReadNoise**2 + (self.metadata.Camera.NoiseFactor**2)*self.metadata.Camera.ElectronsPerCount*self.metadata.Camera.TrueEMGain*np.maximum(dataROI, 1)/nSlices)/self.metadata.Camera.ElectronsPerCount
#phConv = self.metadata.Camera.ElectronsPerCount/self.metadata.Camera.TrueEMGain
#nPhot = dataROI*phConv
if self.noiseSigma is None:
sigma = self._calc_sigma(dataROI)
else:
sigma = self.noiseSigma[xslice, yslice, 0:2]
sigma[:, :, 1] = self.noiseSigma[xslice2, yslice2, 1]
sigma = ndimage.maximum_filter(sigma, [3, 3, 0])
if self.metadata.getOrDefault('Analysis.subtractBackground', True):
#print 'bgs'
if not self.background is None and len(np.shape(self.background)) > 1:
bgROI = self.background[xslice, yslice, 0:2]
bgROI[:, :, 1] = self.background[xslice2, yslice2, 1]
else:
bgROI = np.zeros_like(dataROI) + self.background
else:
bgROI = np.zeros_like(dataROI)
#generate grid to evaluate function on
Xg = vx * (np.mgrid[xslice] + self.roi_offset[0])
Yg = vy * (np.mgrid[yslice] + self.roi_offset[1])
#generate a corrected grid for the red channel
#note that we're cheating a little here - for shifts which are slowly
#varying we should be able to set Xr = Xg + delta_x(\bar{Xr}) and
#similarly for y. For slowly varying shifts the following should be
#equivalent to this. For rapidly varying shifts all bets are off ...
#DeltaX, DeltaY = twoColour.getCorrection(Xg.mean(), Yg.mean(), self.metadata['chroma.dx'],self.metadata['chroma.dy'])
Xr = Xg + DeltaX - vx * dxp
Yr = Yg + DeltaY - vy * dyp
return Xg, Yg, Xr, Yr, dataROI, bgROI, sigma, xslice, yslice, xslice2, yslice2
def tile_pyramid(out_folder, ds, xm, ym, mdh, split=False, skipMoveFrames=False, shiftfield=None,
mixmatrix=[[1., 0.], [0., 1.]],
correlate=False, dark=None, flat=None, pyramid_tile_size=256):
frameSizeX, frameSizeY, numFrames = ds.shape[:3]
if split:
from PYME.Acquire.Hardware import splitter
frameSizeY /= 2
nchans = 2
unmux = splitter.Unmixer(shiftfield, 1e3 * mdh.getEntry('voxelsize.x'))
else:
nchans = 1
#x & y positions of each frame
xps = xm(np.arange(numFrames))
yps = ym(np.arange(numFrames))
if mdh.getOrDefault('CameraOrientation.FlipX', False):
xps = -xps
if mdh.getOrDefault('CameraOrientation.FlipY', False):
yps = -yps
rotate_cam = mdh.getOrDefault('CameraOrientation.Rotate', False)
#give some room at the edges
bufSize = 0
if correlate:
bufSize = 300
x0 = xps.min()
y0 = yps.min()
xps -= x0
yps -= y0
#convert to pixels
xdp = (bufSize + (xps / (mdh.getEntry('voxelsize.x'))).round()).astype('i')
ydp = (bufSize + (yps / (mdh.getEntry('voxelsize.y'))).round()).astype('i')
#calculate a weighting matrix (to allow feathering at the edges - TODO)
weights = np.ones((frameSizeX, frameSizeY, nchans))
#weights[:, :10, :] = 0 #avoid splitter edge artefacts
#weights[:, -10:, :] = 0
#print weights[:20, :].shape
edgeRamp = min(100, int(.25 * ds.shape[0]))
weights[:edgeRamp, :, :] *= np.linspace(0, 1, edgeRamp)[:, None, None]
weights[-edgeRamp:, :, :] *= np.linspace(1, 0, edgeRamp)[:, None, None]
weights[:, :edgeRamp, :] *= np.linspace(0, 1, edgeRamp)[None, :, None]
weights[:, -edgeRamp:, :] *= np.linspace(1, 0, edgeRamp)[None, :, None]
roi_x0, roi_y0 = get_camera_roi_origin(mdh)
ROIX1 = roi_x0 + 1
ROIY1 = roi_y0 + 1
ROIX2 = ROIX1 + mdh.getEntry('Camera.ROIWidth')
ROIY2 = ROIY1 + mdh.getEntry('Camera.ROIHeight')
if dark is None:
offset = float(mdh.getEntry('Camera.ADOffset'))
else:
offset = 0.
P = ImagePyramid(out_folder, pyramid_tile_size, x0=x0, y0=y0, pixel_size=mdh.getEntry('voxelsize.x'))
logger.debug('Adding base tiles ...')
t1 = time.time()
for i in range(int(mdh.getEntry('Protocol.DataStartsAt')), numFrames):
if xdp[i - 1] == xdp[i] or not skipMoveFrames:
x_i = xdp[i]
y_i = ydp[i]
d = ds[:, :, i].astype('f')
if not dark is None:
d = d - dark
if not flat is None:
d = d * flat
if split:
d = np.concatenate(unmux.Unmix(d, mixmatrix, offset, [ROIX1, ROIY1, ROIX2, ROIY2]), 2)
d_weighted = weights * d
# orient frame - TODO - check if we need to flip x and y?!
if rotate_cam:
#print('adding base tile from frame %d [transposed]' % i)
P.add_base_tile(x_i, y_i, d_weighted.T.squeeze(), weights.T.squeeze())
else:
#print('adding base tile from frame %d' % i)
P.add_base_tile(x_i, y_i, d_weighted.squeeze(), weights.squeeze())
t2 = time.time()
logger.debug('Added base tiles in %fs' % (t2 - t1))
#P._occ.flush()
logger.debug(time.time() - t2)
logger.debug('Updating pyramid ...')
P.update_pyramid()
logger.debug(time.time() - t2)
logger.debug('Done')
return P