def test(img):
imgT = TransformView.transformView(img, aff, interval,
MultiViewDeconvolution.minValueImg,
MultiViewDeconvolution.outsideValueImg,
1) # 1: linear interpolation
imgA = ArrayImgs.floats(dimensions)
ImgUtil.copy(ImgView.wrap(imgT, imgA.factory()), imgA)
def classify(img, classifier, class_names, ops=None, distribution_class_index=-1):
""" img: a 2D RandomAccessibleInterval.
classifier: a WEKA Classifier instance, like SMO or FastRandomForest, etc. Any.
If it's a string, interprets it as a file path and attempts to deserialize
a previously saved trained classifier.
class_names: the list of names of each class to learn.
ops: the filter bank of ImgMath ops for the img.
distribution_class_index: defaults to -1, meaning return the class index for each pixel.
When larger than -1, it's interpreted as a class index, and
returns instead the floating-point value of each pixel in
the distribution of that particular class index. """
if type(classifier) == str:
classifier = SerializationHelper.read(classifier)
ops = ops if ops else filterBank(img)
attributes = ArrayList()
for i in xrange(len(ops)):
attributes.add(Attribute("attr-%i" % i))
#for name in classifier.attributeNames()[0][1]:
# attributes.add(Attribute(name))
attributes.add(Attribute("class", class_names))
info = Instances("structure", attributes, 1)
info.setClassIndex(len(attributes) -1)
opImgs = [compute(op).into(ArrayImgs.floats([img.dimension(0), img.dimension(1)])) for op in ops]
cs_opImgs = Views.collapse(Views.stack(opImgs))
result = ArrayImgs.floats([img.dimension(0), img.dimension(1)])
cr = result.cursor()
cop = Views.iterable(cs_opImgs).cursor()
while cr.hasNext():
tc = cop.next()
vector = array((tc.get(i).getRealDouble() for i in xrange(len(opImgs))), 'd')
vector += array([0], 'd')
di = DenseInstance(1.0, vector)
di.setDataset(info) # the list of attributes
if distribution_class_index > -1:
cr.next().setReal(classifier.distributionForInstance(di)[distribution_class_index])
else:
cr.next().setReal(classifier.classifyInstance(di))
return result
def updatePixels(self):
# Copy interval into pixels
view = Views.interval(
Views.extendZero(Views.hyperSlice(self.img3D, 2, self.indexZ)),
self.interval2D)
aimg = ArrayImgs.floats(
self.getPixels(),
[self.interval2D.dimension(0),
self.interval2D.dimension(1)])
ImgUtil.copy(view, aimg)
def testJython(imgred, imggreen, imgblue):
width, height = rgb.dimension(0), rgb.dimension(1)
hb = zeros(width * height, 'f')
sb = zeros(width * height, 'f')
bb = zeros(width * height, 'f')
cred = imgred.cursor()
cgreen = imggreen.cursor()
cblue = imgblue.cursor()
i = 0
while cred.hasNext():
r = cred.next().getRealFloat()
g = cgreen.next().getRealFloat()
b = cblue.next().getRealFloat()
cmax = max(r, g, b)
cmin = min(r, g, b)
bb[i] = cmax / 255.0
if 0 != cmax:
sb[i] = (cmax - cmin) / cmax
# Else leave sb[i] at zero
if 0 == sb[i]:
h = 0
else:
span = cmax - cmin
redc = (cmax - r) / span
greenc = (cmax - g) / span
bluec = (cmax - b) / span
if r == cmax:
h = bluec - greenc
elif g == cmax:
h = 2.0 + redc - bluec
else:
h = 4.0 + greenc - redc
h /= 6.0
if h < 0:
h += 1.0
hb[i] = h
i += 1
hh = ArrayImgs.floats(hb, [width, height])
ss = ArrayImgs.floats(sb, [width, height])
bb = ArrayImgs.floats(bb, [width, height])
return Views.stack(hh, ss, bb)
def read2DImageROI(path, dimensions, interval, pixelType=UnsignedShortType, header=0, byte_order=ByteOrder.LITTLE_ENDIAN):
""" Read a region of interest (the interval) of an image in a file.
Assumes the image is written with the first dimension moving slowest.
path: the file path to the image file.
dimensions: a sequence of integer values e.g. [512, 512, 512]
interval: two sequences of integer values defining the min and max coordinates, e.g.
[[20, 0], [400, 550]]
pixeltype: e.g. UnsignedShortType, FloatType
header: defaults to zero, the number of bytes between the start of the file and the start of the image data.
Supports only these types: UnsignedByteType, UnsignedShortType, FloatType.
Returns an ArrayImg of the given type.
"""
ra = RandomAccessFile(path, 'r')
try:
width, height = dimensions
minX, minY = interval[0]
maxX, maxY = interval[1]
roi_width, roi_height = maxX - minX + 1, maxY - minY + 1
tailX = width - roi_width - minX
#print minX, minY
#print maxX, maxY
#print roi_width, roi_height
size = roi_width * roi_height
n_bytes_per_pixel = pixelType().getBitsPerPixel() / 8
#print n_bytes_per_pixel
bytes = zeros(size * n_bytes_per_pixel, 'b')
# Read only the 2D ROI
ra.seek(header + (minY * width + minX) * n_bytes_per_pixel)
for h in xrange(roi_height):
ra.readFully(bytes, h * roi_width * n_bytes_per_pixel, roi_width * n_bytes_per_pixel)
ra.skipBytes((tailX + minX) * n_bytes_per_pixel)
# Make an image
roiDims = [roi_width, roi_height]
if UnsignedByteType == pixelType:
return ArrayImgs.unsignedBytes(bytes, roiDims)
if UnsignedShortType == pixelType:
shorts = zeros(size, 'h')
ByteBuffer.wrap(bytes).order(byte_order).asShortBuffer().get(shorts)
return ArrayImgs.shorts(shorts, roiDims)
if FloatType == pixelType:
floats = zeros(size, 'f')
ByteBuffer.wrap(bytes).order(byte_order).asFloatBuffer().get(floats)
return ArrayImgs.floats(floats, roiDims)
finally:
ra.close()
def populateInstances(instances, synth_imgs, class_index, mins, maxs):
# Populate the training data: create the filter bank for each feature image
# by reading values from the interval defined by mins and maxs
target = ArrayImgs.floats([width, height])
interval = FinalInterval(mins, maxs)
n_samples = Intervals.numElements(interval)
for img in synth_imgs:
vectors = [zeros(len(attributes), 'd') for _ in xrange(n_samples)]
for k, op in enumerate(filterBank(img, sumType=DoubleType())):
imgOp = compute(op).into(target)
for i, v in enumerate(Views.interval(imgOp, interval)):
vectors[i][k] = v.getRealDouble()
for vector in vectors:
vector[-1] = class_index
instances.add(DenseInstance(1.0, vector))
def readFloats(path, dimensions, header=0, byte_order=ByteOrder.LITTLE_ENDIAN):
""" Read a file as an ArrayImg of FloatType """
size = reduce(operator.mul, dimensions)
ra = RandomAccessFile(path, 'r')
try:
if header < 0:
# Interpret from the end: useful for files with variable header lengths
# such as some types of uncompressed TIFF formats
header = ra.length() + header
ra.skipBytes(header)
bytes = zeros(size * 4, 'b')
ra.read(bytes)
floats = zeros(size, 'f')
ByteBuffer.wrap(bytes).order(byte_order).asFloatBuffer().get(floats)
return ArrayImgs.floats(floats, dimensions)
finally:
ra.close()
def twoStep(index=0):
# The current way:
img = klb.readFull(filepaths[index]) # klb_loader.get(filepaths[index])
imgE = Views.extendZero(img)
imgI = Views.interpolate(imgE, NLinearInterpolatorFactory())
imgT = RealViews.transform(imgI, cmIsotropicTransforms[index])
imgB = Views.zeroMin(Views.interval(imgT, roi[0],
roi[1])) # bounded: crop with ROI
imgBA = ArrayImgs.unsignedShorts(Intervals.dimensionsAsLongArray(imgB))
ImgUtil.copy(ImgView.wrap(imgB, imgBA.factory()), imgBA)
imgP = prepareImgForDeconvolution(
imgBA,
affine3D(fineTransformsPostROICrop[index]).inverse(),
FinalInterval([0, 0, 0], [imgB.dimension(d) - 1 for d in xrange(3)]))
# Copy transformed view into ArrayImg for best performance in deconvolution
imgA = ArrayImgs.floats(Intervals.dimensionsAsLongArray(imgP))
ImgUtil.copy(ImgView.wrap(imgP, imgA.factory()), imgA)
IL.wrap(imgA, "two step").show()
def prepare(index):
# Prepare the img for deconvolution:
# 0. Transform in one step.
# 1. Ensure its pixel values conform to expectations (no zeros inside)
# 2. Copy it into an ArrayImg for faster recurrent retrieval of same pixels
syncPrint("Preparing %s CM0%i for deconvolution" % (tm_dirname, index))
img = klb_loader.get(filepaths[index]) # of UnsignedShortType
imgP = prepareImgForDeconvolution(
img, transforms[index], target_interval) # returns of FloatType
# Copy transformed view into ArrayImg for best performance in deconvolution
imgA = ArrayImgs.floats(Intervals.dimensionsAsLongArray(imgP))
#ImgUtil.copy(ImgView.wrap(imgP, imgA.factory()), imgA)
ImgUtil.copy(imgP, imgA, n_threads / 2) # parallel copying
syncPrint("--Completed preparing %s CM0%i for deconvolution" %
(tm_dirname, index))
imgP = None
img = None
return (index, imgA)
def oneStep(index=0):
# Combining transforms into one, via a translation to account of the ROI crop
img = klb.readFull(filepaths[index]) # klb_loader.get(filepaths[index])
t1 = cmIsotropicTransforms[index]
t2 = affine3D(
[1, 0, 0, -roi[0][0], 0, 1, 0, -roi[0][1], 0, 0, 1, -roi[0][2]])
t3 = affine3D(fineTransformsPostROICrop[index]).inverse()
aff = AffineTransform3D()
aff.set(t1)
aff.preConcatenate(t2)
aff.preConcatenate(t3)
# Final interval is now rooted at 0,0,0 given that the transform includes the translation
imgP = prepareImgForDeconvolution(
img, aff,
FinalInterval([0, 0, 0],
[maxC - minC for minC, maxC in izip(roi[0], roi[1])]))
# Copy transformed view into ArrayImg for best performance in deconvolution
imgA = ArrayImgs.floats(Intervals.dimensionsAsLongArray(imgP))
ImgUtil.copy(ImgView.wrap(imgP, imgA.factory()), imgA)
IL.wrap(imgA, "one step index %i" % index).show()
def createTrainingData(img, samples, class_names, n_samples=0, ops=None):
""" img: a 2D RandomAccessibleInterval.
samples: a sequence of long[] (or int numeric sequence or Localizable) and class_index pairs; can be a generator.
n_samples: optional, the number of samples (in case samples is e.g. a generator).
class_names: a list of class names, as many as different class_index.
ops: optional, the sequence of ImgMath ops to apply to the img, defaults to filterBank(img)
return an instance of WEKA Instances
"""
ops = ops if ops else filterBank(img)
if 0 == n_samples:
n_samples = len(samples)
# Define a WEKA Attribute for each feature (one for op in the filter bank, plus the class)
attribute_names = ["attr-%i" % (i+1) for i in xrange(len(ops))]
attributes = ArrayList()
for name in attribute_names:
attributes.add(Attribute(name))
# Add an attribute at the end for the classification classes
attributes.add(Attribute("class", class_names))
# Create the training data structure
training_data = Instances("training", attributes, n_samples)
training_data.setClassIndex(len(attributes) -1)
opImgs = [compute(op).into(ArrayImgs.floats([img.dimension(0), img.dimension(1)])) for op in ops]
ra = Views.collapse(Views.stack(opImgs)).randomAccess()
for position, class_index in samples:
ra.setPosition(position)
tc = ra.get()
vector = array((tc.get(i).getRealDouble() for i in xrange(len(opImgs))), 'd')
vector += array([class_index], 'd')
training_data.add(DenseInstance(1.0, vector))
return training_data
from net.imglib2.type.numeric.real import FloatType ft = FloatType(10) ft.pow(2) print ft from net.imglib2.algorithm.math.ImgMath import compute, add, power from net.imglib2.img.array import ArrayImgs from net.imglib2.img.display.imagej import ImageJFunctions as IL img = ArrayImgs.floats([10, 10, 10]) compute(add(img, 5)).into(img) # in place compute(power(img, 2)).into(img) # in place print 25 * 10 * 10 * 10 == sum(t.get() for t in img.cursor()) IL.wrap(img, "5 squared").show()
from net.imglib2.algorithm.math import ImgMath
from net.imglib2.util import ImgUtil
from net.imglib2.img import ImgView
import sys
sys.path.append("/home/albert/lab/scripts/python/imagej/IsoView-GCaMP/")
from lib.util import timeit
roi = (
[1, 228, 0], # top-left coordinates
[1 + 406 - 1, 228 + 465 - 1,
0 + 325 - 1]) # bottom-right coordinates (inclusive, hence the -1)
dimensions = [maxC - minC + 1 for minC, maxC in zip(roi[0], roi[1])]
imgU = ArrayImgs.unsignedShorts(dimensions)
imgF = ArrayImgs.floats(dimensions)
#c = imgF.cursor()
#while c.hasNext():
# c.next().set(random() * 65535)
ImgMath.compute(ImgMath.number(17)).into(imgF)
ImgMath.compute(ImgMath.img(imgF)).into(imgU)
aff = AffineTransform3D()
"""
aff.set(1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0)
"""
aff.set(*[
0.9999949529841275, -0.0031770224721305684, 2.3118912942710207e-05,
-1.6032353998500826, 0.003177032139125933, 0.999994860398559,
-0.00043086338151948394, -0.4401520585103873, -2.1749931475206362e-05,
for i, v in enumerate(Views.interval(imgOp, interval)):
vectors[i][k] = v.getRealDouble()
for vector in vectors:
vector[-1] = class_index
instances.add(DenseInstance(1.0, vector))
# pick pixels on the black line for class 0 (membrane), 4x4
populateInstances(training_data, synth_imgs_membrane, 0, [14, 14], [17, 17])
# pick pixels in the very center for class 1 (mitochondrial boundary), 2x2
populateInstances(training_data, synth_imgs_mit_boundary, 1, [15, 15],
[16, 16])
# Populate the training data for class "other" from two images
# entirely filled with background or foreground plus noise
target = ArrayImgs.floats([width, height])
interval = FinalInterval([14, 14], [17, 17])
n_samples = Intervals.numElements(interval)
for ci, v in enumerate([fillValue, backgroundValue]):
for _ in xrange(training_data.size() /
4): # the other 2/4 are the membrane and mit boundary
other = syntheticEM([], width, height, 0, v, noise=True)
vectors = [zeros(len(attributes), 'd') for _ in xrange(n_samples)]
for k, op in enumerate(filterBank(IL.wrap(other),
sumType=DoubleType())):
imgOp = compute(op).into(target)
for i, v in enumerate(Views.interval(imgOp, interval)):
vectors[i][k] = v.getRealDouble()
for vector in vectors:
vector[-1] = ci + 2 # class index
training_data.add(DenseInstance(1.0, vector))
# @OpService ops # @float[] arr # @long[] dims # @OUTPUT Img out from net.imagej.ops import Ops from net.imglib2.img import Img from net.imglib2.img.array import ArrayImgs arr = list(arr) dims = list(dims) out = ArrayImgs.floats(arr, dims)
def __init__(self, dimensions, filenames):
super(Thread, self).__init__()
self.filenames = filenames
self.aimg = ArrayImgs.floats(dimensions)
self.klb = KLB.newInstance()
def factory(self):
return None
def copy(self):
return self # stateless, so safe
def randomAccess(
self,
interval=None
): # optional argument handles case of having randomAccess() and randomAccess(interval).
return self.cursor()
def cursor(self):
return FnCursor(self.numDimensions(), -pi / 2, self.dimension(0) / 4)
def localizingCursor(self):
return self.cursor()
img = FnImg([512, 512])
IL.show(img) # shows black, yet above the get() prints values
aimg = ArrayImgs.floats([512, 512])
c1 = img.cursor()
c2 = aimg.cursor()
while c2.hasNext():
t = c2.next()
c1.setPosition(c2)
c2.next().set(c1.get())
IL.show(aimg) # Shows black, yet above the get() prints values
imgB0,
transformedView(imgB1, matrices["imgB0-imgB1"]),
transformedView(imgB2, matrices["imgB0-imgB2"]),
transformedView(imgB3, matrices["imgB0-imgB3"])
]
#viewInBDV(*transformed)
viewAsStack(*transformed)
exe = newFixedThreadPool(4)
try:
# Copy into ArrayImg
def copyIntoArrayImg(img):
return ImgMath.compute(ImgMath.img(img)).into(ArrayImgs.floats([img.dimension(d) for d in xrange(img.numDimensions())]))
futures = [exe.submit(Task(copyIntoArrayImg, img)) for img in [imgB0, imgB1, imgB2, imgB3]]
imgB0, imgB1, imgB2, imgB3 = [f.get() for f in futures]
finally:
exe.shutdown()
viewAsStack(imgB0, imgB1, imgB2, imgB3) # ArrayImg instances
# Read the kernel as a FloatType ArrayImg
kernel = readFloats("/home/albert/lab/Raghav-IsoView-PSF/PSF-19x19x25.tif", [19, 19, 25], header=434)
def affine3D(matrix):
aff = AffineTransform3D()
aff.set(*matrix)
return aff
def readKernel(path):
if kernel_header is None:
imp = IJ.openImage(path)
return ArrayImgs.floats(imp.getProcessor().getPixels(),
[imp.getWidth(), imp.getHeight(), imp.getNSlices()])
return readFloats(path, kernel_dimensions, kernel_header)
def testASMFloats():
img2 = ArrayImgs.floats(dimensions)
ImgUtil.copy(
ImgView.wrap(
Converters.convertRandomAccessibleIterableInterval(
img1, sampler_conv_floats), img1.factory()), img2)
from net.imglib2.img.array import ArrayImgs
import sys
sys.path.append("/home/albert/lab/scripts/python/imagej/IsoView-GCaMP/")
from random import random
from net.imglib2.loops import LoopBuilder
from net.imglib2.type.numeric.real import FloatType
from lib.loop import createBiConsumerTypeSet, createBiConsumerTypeSet2, binaryLambda, nthLambda
from java.util.function import BiConsumer
img1 = ArrayImgs.floats([10, 10, 10])
cursor = img1.cursor()
for t in img1:
t.setReal(random())
img2 = ArrayImgs.floats([10, 10, 10])
#copyIt = createBiConsumerTypeSet(FloatType) # works well
#copyIt = createBiConsumerTypeSet2(FloatType) # works well
#LoopBuilder.setImages(img1, img2).forEachPixel(copyIt)
# Works well:
#copyIt = binaryLambda(FloatType, "set", FloatType,
# interface=BiConsumer, interface_method="accept")
copyIt = nthLambda(FloatType, "set", [FloatType], BiConsumer, "accept")
ra1 = img1.randomAccess()
ra2 = img2.randomAccess()
pos = [3, 2, 5]
ra1.setPosition(pos)
def test(red, green, blue, easy=True):
saturation = let(
"red", red, "green", green, "blue", blue, "max",
maximum("red", "green", "blue"), "min",
minimum("red", "green", "blue"),
IF(EQ(0, "max"), THEN(0), ELSE(div(sub("max", "min"), "max"))))
brightness = div(maximum(red, green, blue), 255.0)
hue = IF(
EQ(0, saturation), THEN(0),
ELSE(
let(
"red", red, "green", green, "blue", blue, "max",
maximum("red", "green", "blue"), "min",
minimum("red", "green", "blue"), "range", sub("max", "min"),
"redc", div(sub("max", "red"), "range"), "greenc",
div(sub("max", "green"), "range"), "bluec",
div(sub("max", "blue"), "range"), "hue",
div(
IF(
EQ("red", "max"), THEN(sub("bluec", "greenc")),
ELSE(
IF(EQ("green", "max"),
THEN(sub(add(2, "redc"), "bluec")),
ELSE(sub(add(4, "greenc"), "redc"))))), 6),
IF(LT("hue", 0), THEN(add("hue", 1)), ELSE("hue")))))
#print hierarchy(hue)
#print "hue view:", hue.view( FloatType() ).iterationOrder()
if easy:
# About 26 ms
"""
hsb = Views.stack( hue.view( FloatType() ),
saturation.view( FloatType() ),
brightness.view( FloatType() ) )
"""
# About 13 ms: half! Still much worse than plain ImageJ,
# but the source images are iterated 4 times, rather than just once,
# and the saturation is computed twice,
# and the min, max is computed 3 and 4 times, respectively.
hsb = Views.stack(hue.viewDouble(FloatType()),
saturation.viewDouble(FloatType()),
brightness.viewDouble(FloatType()))
"""
# Even worse: ~37 ms
width, height = rgb.dimension(0), rgb.dimension(1)
h = compute(hue).into(ArrayImgs.floats([width, height]))
s = compute(saturation).into(ArrayImgs.floats([width, height]))
b = compute(brightness).into(ArrayImgs.floats([width, height]))
hsb = Views.stack( h, s, b )
"""
imp = IL.wrap(hsb, "HSB view")
else:
# Tested it: takes more time (~40 ms vs 26 ms above)
width, height = rgb.dimension(0), rgb.dimension(1)
hb = zeros(width * height, 'f')
sb = zeros(width * height, 'f')
bb = zeros(width * height, 'f')
h = ArrayImgs.floats(hb, [width, height])
s = ArrayImgs.floats(sb, [width, height])
b = ArrayImgs.floats(bb, [width, height])
#print "ArrayImg:", b.iterationOrder()
ImgUtil.copy(ImgView.wrap(hue.view(FloatType()), None), h)
ImgUtil.copy(ImgView.wrap(saturation.view(FloatType()), None), s)
ImgUtil.copy(ImgView.wrap(brightness.view(FloatType()), None), b)
stack = ImageStack(width, height)
stack.addSlice(FloatProcessor(width, height, hb, None))
stack.addSlice(FloatProcessor(width, height, sb, None))
stack.addSlice(FloatProcessor(width, height, bb, None))
imp = ImagePlus("hsb", stack)
return imp
imp = IJ.getImage() dimensions = [imp.getWidth(), imp.getHeight()] ip = imp.getProcessor() pixels = ip.getPixels() # In practice, you never want to do this below, # and instead you'd use the built-in wrapper: ImageJFunctions.wrap(imp) # This is merely for illustration of how to use ArrayImgs with an existing pixel array if isinstance(ip, ByteProcessor): img1 = ArrayImgs.unsignedBytes(pixels, dimensions) elif isinstance(ip, ShortProcessor): img1 = ArrayImgs.unsignedShorts(pixels, dimensions) elif isinstance(ip, FloatProcessor): img1 = ArrayImgs.floats(pixels, dimensions) else: print "Can't handle image of type:", type(ip).getName() # An empty image of float[] img2 = ArrayImgs.floats(dimensions) # View it as RandomAccessibleInterval<FloatType> by converting on the fly # using a generic RealType to FloatType converter floatView = Converters.convertRAI(img1, RealFloatConverter(), FloatType()) # The above 'floatView' can be used as an image: one that gets always converted on demand. # If you only have to iterate over the pixels just once, there's no need to create a new image. IL.show(floatView, "32-bit view of the 8-bit")
imgE = Views.extendZero(target)
# Integrate every dimension, cummulatively by writing into
# a target image that is also the input
for d in xrange(img.numDimensions()):
coord = [0] * img.numDimensions() # array of zeros
coord[d] = -1
# Cummulative sum along the current dimension
# Note that instead of the ImgMath offset op,
# we could have used Views.translate(Views.extendZero(target), [1, 0]))
# (Notice though the sign change in the translation)
integral = add(target, offset(imgE, coord))
compute(integral).into(target)
# The target is the integral image
integralImg = target
# Read out blocks of radius 5 (i.e. 10x10 for a 2d image)
# in a way that is entirely n-dimensional (applies to 1d, 2d, 3d, 4d ...)
radius = 5
nd = img.numDimensions()
op = div(block(Views.extendBorder(integralImg), [radius] * nd),
pow(radius * 2, nd)) # divide by total number of pixels in the block
blurred = ArrayImgs.floats(Intervals.dimensionsAsLongArray(img))
compute(op).into(blurred, FloatType())
# Show the blurred image with the same LUT as the original
imp2 = IL.wrap(blurred, "integral image radius 5 blur")
imp2.getProcessor().setLut(imp.getProcessor().getLut())
imp2.show()