def pureConcaternation(probPath, rawPath):
"""concatanating int with the raw image """
assert len(splitPath(probPath)) == 2, "file " + probPath + "seems to not have the .h5 extension"
print 'load prob data'
data = vigra.readHDF5(*splitPath(probPath)).squeeze()
if len(data.shape) != 4:
print "WARNING: untested for data other than 3d + channel."
# if isinstance(data,vigra.VigraArray):
# data = data.view(np.ndarray)
print 'normalize prob data'
# normalize probabilities and save them as hdf5
data = np.require(data, dtype=np.float32)
data -= data.min()
data *= 255 / data.max()
data = data.astype(np.uint8)
if rawPath != None:
print 'load raw data'
assert len(splitPath(rawPath)) == 2, "file " + rawPath + "seems to not have the .h5 extension"
raw = vigra.readHDF5(*splitPath(rawPath)).squeeze()
print 'concaternate'
embed()
data = np.concatenate((raw[:,:,:,None], data), axis=3)
vigra.writeHDF5(data.astype(np.uint8), probPath.split('.h5')[0] + '_stacked.h5', splitPath(probPath)[1])
def upsample(probPath, rawPath, transpose, channel, targetShape=None, exportPath=None):
""" upsampling of multichannel data of an ilp (ignoring the last dimension which should be used for channels) and concatanating int with the raw image """
splitProbPath = probPath.split('.h5')
assert len(splitProbPath) !=1, "file " + splitProbPath[0] + "seems to not have the .h5 extension"
data = vigra.readHDF5(splitProbPath[0] + '.h5', splitProbPath[1]).squeeze()
if transpose['prob']:
data = np.transpose(data, [2,1,0,3])
if channel is not None:
axisTags = data.axistags
data = data[:,:,:,[channel]]
data.axistags = axisTags
if len(data.shape) != 4:
print "WARNING: untested for data other than 3d + channel."
# if isinstance(data,vigra.VigraArray):
# data = data.view(np.ndarray)
if rawPath != None:
splitRawPath = rawPath.split('.h5')
assert len(splitRawPath) !=1, "file " + splitRawPath[0] + "seems to not have the .h5 extension"
raw = vigra.readHDF5(splitRawPath[0] + '.h5', splitRawPath[1]).squeeze()
if transpose['raw']:
raw = raw.T
targetShape = raw.shape
print 'check please: theoretical', [size*2 for size in data.shape[:-1]], 'actual shape', targetShape
elif targetShape is not None:
print 'check please: theoretical', [size*2 for size in data.shape[:-1]], 'actual shape', targetShape
else:
targetShape = [size*2 for size in data.shape[:-1]]
embed()
# normalize probabilities and save them as hdf5
data = np.require(data, dtype=np.float32)
data = vigra.sampling.resize(data, shape=targetShape, order=2)
data -= data.min()
data *= 255 / data.max()
data = data.astype(np.uint8)
if rawPath != None:
data = np.concatenate((raw[:,:,:,None], data), axis=3)
if exportPath is None:
exportPath = splitProbPath[0] + '_upsampled.h5/' + splitProbPath[1]
print 'exporting to ', exportPath
embed()
vigra.writeHDF5(data.astype(np.uint8), *splitPath(exportPath))
def downsampleProject(ilpPath, adjustScales=False, resizeRaw=True):
assert ilpPath.endswith('.ilp'), 'ilp file is required for downsampling. did you mean to upsample?'
name = ilpPath.split('.ilp')[0] + '_small.ilp'
print 'copy original to', name
copyfile(ilpPath, name)
f = File(name)
# move feature matrix to smaller sigmas (first row should stay the same, second should not be overwritten, but 'or'ed, the rest is just moved to the next smaller sigma.
if adjustScales:
matrix = f['FeatureSelections/SelectionMatrix/'].value
matrix[:,1] = np.logical_or(matrix[:,2], matrix[:,1])
matrix[:,2:-1] = matrix[:,3:]
matrix[:,-1] = False
del f['FeatureSelections/SelectionMatrix/']
f['FeatureSelections/SelectionMatrix/'] = matrix
for key in f['Input Data/infos'].keys():
path = splitPath(f['Input Data/infos/' + key + '/Raw Data/filePath'].value)
# resize the labels
labels = labelManager(name)
labeledBlocks = labels.getSubBlocks()
labeledBlocksResized = list()
for labelsOffset, labelsData in labeledBlocks:
# last dimension is singelton.
labelsData = labelsData[...,0]
originalShape = labelsData.shape
# memorize the final target shape.
targetShape = [max(int(ordinate/2), 1) for ordinate in originalShape]
# size must be at least 4x4x4 to work without an error.
if np.any(np.array(originalShape) < np.array([4,4,4])):
shapeBeforeScaling = [max(4, ordinate) for ordinate in originalShape]
dataBeforeScaling = np.zeros(shapeBeforeScaling, dtype=np.uint8)
slicing = [slice(0, size) for size in originalShape]
dataBeforeScaling[slicing] = labelsData
labelsData = dataBeforeScaling
originalShape = shapeBeforeScaling
shapeAfterResize = [max(int(ordinate/2), 1) for ordinate in originalShape]
# resize the labels
resized = np.zeros(shapeAfterResize, dtype=np.uint8)
vigra.graphs.downsampleLabels(labelsData, int(labelsData.max()), 0.05, resized) # get back the original shape
resized = resized[[slice(0,size) for size in targetShape]]
offsetResized = tuple([int(ordinate/2) for ordinate in labelsOffset])
labeledBlocksResized.append((offsetResized, resized))
# clear the old labelds
labels.clear()
labels.flush()
for labeledBlock in labeledBlocksResized:
labels.addBlockLabel(labeledBlock[1][:,:,:,None], labeledBlock[0])
labels.flush()
if resizeRaw:
print 'resize raw data in', path
rawData = vigra.readHDF5(*path)
resizedShape = [int(ordinate/2) for ordinate in rawData.shape]
exportRawPath = [path[0].split('.')[0] + '_resized.h5', path[1]]
rawSmall = vigra.sampling.resize(rawData.astype(np.float32), resizedShape , 0)
vigra.writeHDF5(rawSmall, *exportRawPath)
try:
del f['Input Data/infos/' + key + '/Raw Data/filePath']
except KeyError:
pass
f['Input Data/infos/' + key + '/Raw Data/filePath'] = exportRawPath[0] + '/' + exportRawPath[1]
parser.add_argument("--ignoreRaw", default=False, action="store_true")
parser.add_argument("--rawT", default=False, action="store_true")
parser.add_argument("--probT", default=False, action="store_true")
parser.add_argument("--channel", default=None, type=int)
parser.add_argument("--shapeFromRaw", default=None, type=str)
parser.add_argument("--exportPath", default=None, type=str)
args = parser.parse_args()
path = args.input_path
transpose = dict()
transpose['raw'] = args.rawT
transpose['prob'] = args.probT
if args.shapeFromRaw is not None:
targetShape = vigra.readHDF5(*splitPath(args.shapeFromRaw)).shape
if transpose['raw']:
targetShape = targetShape[::-1]
print 'targetShape set to', targetShape
else:
targetShape = None
if args.upsample:
upsample(path, args.stackWithRaw, transpose, args.channel, targetShape, args.exportPath)
elif args.stackWithRaw is not None:
pureConcaternation(args.input_path, args.stackWithRaw)
else:
downsampleProject(path, args.adjustScales, not args.ignoreRaw)
# upsampleMultiChannelData('/home/timo/multiscaletest/results/actualHoles_resized_probs.h5', 'exported_data')
