def omni_generator(f_input, begin=0, end=64, step=64):
for i in range(begin, end, step):
if i + step > end:
data = dxchange.read_tiff_stack(f_input, ind=range(i, end))
else:
data = dxchange.read_tiff_stack(f_input, ind=range(i, i + step))
yield (i, data)
def omni_read(f_input, begin=None, end=None):
'''support tiff, tiff stack, hdf5'''
if not f_input:
return None
f_input = os.path.abspath(f_input)
matches = re.match(
r'^(?P<dirname>.*)/(?P<fname>.*)\.(?P<ext>[^:]*)($|:(?P<dataset>.*$))',
f_input)
# print(matches.groupdict())
if matches['ext'] == 'tif' or matches['ext'] == 'tiff':
if begin is not None and end is not None:
data = dxchange.read_tiff_stack(f_input, ind=range(begin, end))
else:
# print(f_input)
S, L = check_stack_len(f_input)
# print(S,L)
if L > 1:
data = dxchange.read_tiff_stack(f_input, ind=range(S, S + L))
else:
data = dxchange.read_tiff(f_input)
elif matches['ext'] == 'h5' or matches['ext'] == 'hdf5':
tokens = f_input.split(':')
dataset_name = tokens[1]
f_input = h5py.File(tokens[0], 'r')
data = np.asarray(f_input[tokens[1]])
else:
print('not implemented file type')
return data
def get_reslice(recon_folder,
chunk_size=50,
slice_y=None,
slice_x=None,
rotate=0):
filelist = glob.glob(os.path.join(recon_folder, 'recon*.tiff'))
inds = []
digit = None
for i in filelist:
i = os.path.split(i)[-1]
regex = re.compile(r'\d+')
a = regex.findall(i)[0]
if digit is None:
digit = len(a)
inds.append(int(a))
chunks = []
chunk_st = np.min(inds)
chunk_end = chunk_st + chunk_size
sino_end = np.max(inds) + 1
while chunk_end < sino_end:
chunks.append((chunk_st, chunk_end))
chunk_st = chunk_end
chunk_end += chunk_size
chunks.append((chunk_st, sino_end))
a = dxchange.read_tiff_stack(filelist[0], range(chunks[0][0],
chunks[0][1]), digit)
if rotate != 0:
a = scipy.ndimage.interpolation.rotate(a,
rotate,
axes=(1, 2),
reshape=False)
if slice_y is not None:
slice = a[:, slice_y, :]
elif slice_x is not None:
slice = a[:, ::-1, slice_x]
else:
raise ValueError('Either slice_y or slice_x must be specified.')
for (chunk_st, chunk_end) in chunks[1:]:
a = dxchange.read_tiff_stack(filelist[0], range(chunk_st, chunk_end),
digit)
if rotate != 0:
a = scipy.ndimage.interpolation.rotate(a,
rotate,
axes=(1, 2),
reshape=False)
if slice_y is not None:
slice = np.append(slice, a[:, slice_y, :], axis=0)
elif slice_x is not None:
slice = np.append(slice, a[:, ::-1, slice_x], axis=0)
else:
raise ValueError('Either slice_y or slice_x must be specified.')
return slice
def load_raw(top, index_start):
"""
Function description.
Parameters
----------
parameter_01 : type
Description.
parameter_02 : type
Description.
parameter_03 : type
Description.
Returns
-------
return_01
Description.
"""
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
return rdata
def load_raw(top, index_start):
"""
Load a stack of tiff images.
Parameters
----------
top : str
Top data directory.
index_start : int
Image index start.
Returns
-------
ndarray
3D stack of images.
"""
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
return rdata
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=1, help="index of the first image: 10001 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
particle_bed_reference = particle_bed_location(rdata[0], plot=False)
print("Particle bed location: ", particle_bed_reference)
print("Laser on?: ", laser_on(rdata, particle_bed_reference))
print("Shutter closed on image: ", shutter_off(rdata))
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=0,
help="index of the first image: 100 (default 0)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
slider(rdata)
def execute(self, startswith, seperator, idx, output_dataobj):
if (startswith is None) or (seperator is None) or (idx is None):
self.LogError('Please enter scanning requirements for files')
return False
filelist = os.listdir(self.ipath)
filelist = list(
filter(lambda filename: filename.startswith(startswith), filelist))
if len(filelist) == 0:
self.LogError('Can not scan required files, Please check!')
return False
else:
filelist.sort()
self.LogInfo('First Input File: ' + filelist[0])
if not (filelist[0].split('.')[-1].upper() == 'TIF'):
self.LogError('Files should be TIF')
return False
fileidxs = list(
map(
lambda filename: '.'.join(filename.split('.')[:-1]).split(
seperator)[idx], filelist))
if len(fileidxs) == 0:
self.LogError('Can not scan required files, Please check!')
return False
else:
self.LogInfo('Get ' + str(len(fileidxs)) + ' files in the Path ' +
self.ipath)
self.data[output_dataobj] = dxchange.read_tiff_stack(self.ipath + '/' +
filelist[0],
ind=fileidxs)
self.LogInfo('Load data ' + output_dataobj + ' from ' + self.ipath)
return True
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=0, type=int, default=0, help="index of the first image: 10001 (default 0)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), 'sdat*'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
print(fname, ind_tomo)
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
sdata = rdata
for index in ind_tomo:
print(index)
sdata[index] = rdata[index].byteswap()
#rdata[0] = rdata[0].byteswap()
print(sdata.shape)
slider(sdata)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=0, help="index of the first image: 100 (default 0)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
# View the data
slider(rdata)
# Apply the sobel filter
ndata = scale_to_one(rdata)
ndata = sobel_stack(ndata)
slider(ndata)
blur_radius = 3.0
threshold = .04
nddata = label(ndata, blur_radius, threshold)
slider(ndata)
def large_data_generator(stack_name,
begin=0,
end=64,
step=64,
dtype=None,
multi=False):
for i in range(begin, end, step):
if i + step > end:
data = dxchange.read_tiff_stack(stack_name, ind=range(i, end))
else:
data = dxchange.read_tiff_stack(stack_name, ind=range(i, i + step))
if not multi and dtype == 'uint32':
data = np.nan_to_num(data > 0)
if dtype:
data = data.astype(dtype)
data = np.moveaxis(data, 0, 2)
yield (i, data)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=1,
help="index of the first image: 10001 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
particle_bed_reference = particle_bed_location(rdata[0], plot=False)
print("Particle bed location: ", particle_bed_reference)
# Cut the images to remove the particle bed
cdata = rdata[:, 0:particle_bed_reference, :]
# Find the image when the shutter starts to close
dark_index = shutter_off(rdata)
print("shutter closes on image: ", dark_index)
# Set the [start, end] index of the blocked images, flat and dark.
flat_range = [0, 1]
data_range = [48, dark_index]
dark_range = [dark_index, nfile]
# # for fast testing
# data_range = [48, dark_index]
flat = cdata[flat_range[0]:flat_range[1], :, :]
proj = cdata[data_range[0]:data_range[1], :, :]
dark = np.zeros(
(dark_range[1] - dark_range[0], proj.shape[1], proj.shape[2]))
# if you want to use the shutter closed images as dark uncomment this:
#dark = cdata[dark_range[0]:dark_range[1], :, :]
ndata = tomopy.normalize(proj, flat, dark)
ndata = tomopy.normalize_bg(ndata, air=ndata.shape[2] / 2.5)
ndata = tomopy.minus_log(ndata)
sharpening(ndata)
slider(ndata)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=1, help="index of the first image: 1000 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[0]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
print (nfile, index_start, index_end, fname)
# Select the sinogram range to reconstruct.
start = 0
end = 512
sino=(start, end)
# Read the tiff raw data.
ndata = dxchange.read_tiff_stack(fname, ind=ind_tomo, slc=(sino, None))
# Normalize to 1 using the air counts
ndata = tomopy.normalize_bg(ndata, air=5)
# Set data collection angles as equally spaced between 0-180 degrees.
theta = tomopy.angles(ndata.shape[0])
ndata = tomopy.minus_log(ndata)
# Set binning and number of iterations
binning = 8
iters = 21
print("Original", ndata.shape)
ndata = tomopy.downsample(ndata, level=binning, axis=1)
# ndata = tomopy.downsample(ndata, level=binning, axis=2)
print("Processing:", ndata.shape)
fdir = 'aligned' + '/noblur_iter_' + str(iters) + '_bin_' + str(binning)
print(fdir)
cprj, sx, sy, conv = alignment.align_seq(ndata, theta, fdir=fdir, iters=iters, pad=(10, 10), blur=False, save=True, debug=True)
np.save(fdir + '/shift_x', sx)
np.save(fdir + '/shift_y', sy)
# Write aligned projections as stack of TIFs.
dxchange.write_tiff_stack(cprj, fname=fdir + '/radios/image')
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=1, help="index of the first image: 1000 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[0]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
print (nfile, index_start, index_end, fname)
# Select the sinogram range to reconstruct.
start = 0
end = 512
sino=(start, end)
# Read the tiff raw data.
ndata = dxchange.read_tiff_stack(fname, ind=ind_tomo, slc=(sino, None))
print(ndata.shape)
binning = 8
ndata = tomopy.downsample(ndata, level=binning, axis=1)
print(ndata.shape)
# Normalize to 1 using the air counts
ndata = tomopy.normalize_bg(ndata, air=5)
## slider(ndata)
# Set data collection angles as equally spaced between 0-180 degrees.
theta = tomopy.angles(ndata.shape[0])
rot_center = 960
print("Center of rotation: ", rot_center)
ndata = tomopy.minus_log(ndata)
# Reconstruct object using Gridrec algorithm.
rec = tomopy.recon(ndata, theta, center=rot_center, algorithm='gridrec')
# Mask each reconstructed slice with a circle.
rec = tomopy.circ_mask(rec, axis=0, ratio=0.95)
# Write data as stack of TIFs.
dxchange.write_tiff_stack(rec, fname='/local/dataraid/mark/rec/recon')
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=1,
help="index of the first image: 1000 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[0]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
print(nfile, index_start, index_end, fname)
# Select the sinogram range to reconstruct.
start = 70
end = 72
sino = (start, end)
# Read the tiff raw data.
ndata = dxchange.read_tiff_stack(fname, ind=ind_tomo, slc=(sino, None))
# Set data collection angles as equally spaced between 0-180 degrees.
theta = tomopy.angles(ndata.shape[0])
rot_center = 251
print("Center of rotation: ", rot_center)
#ndata = tomopy.minus_log(ndata)
# Reconstruct object using Gridrec algorithm.
rec = tomopy.recon(ndata, theta, center=rot_center, algorithm='gridrec')
# Mask each reconstructed slice with a circle.
rec = tomopy.circ_mask(rec, axis=0, ratio=0.95)
# Write data as stack of TIFs.
dxchange.write_tiff_stack(rec, fname='/local/dataraid/mark/rec/recon')
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=1,
help="index of the first image: 10001 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
index_end = 10
ind_tomo = range(index_start, index_end)
fname = top + template
print(nfile, index_start, index_end, fname)
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
ndata = rdata
# slider(ndata)
# ndata = sobel_stack(rdata)
# print(ndata[0, :, :])
# slider(ndata)
# blur_radius = 2
# threshold = 0.0000001
# ndata = label(rdata, blur_radius, threshold)
# slider(ndata)
idata = se.rescale_intensity(ndata[0, :, :], out_range=(0, 256))
distance = ndi.distance_transform_edt(idata)
local_maxi = sf.peak_local_max(distance,
labels=idata,
footprint=np.ones((3, 3)),
indices=False)
markers = ndi.label(local_maxi)[0]
labels = seg.watershed(-distance, markers, mask=idata)
slider(labels)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=1, help="index of the first image: 10001 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
particle_bed_reference = particle_bed_location(rdata[0], plot=False)
print("Particle bed location: ", particle_bed_reference)
# Cut the images to remove the particle bed
cdata = rdata[:, 0:particle_bed_reference, :]
# Find the image when the shutter starts to close
dark_index = shutter_off(rdata)
print("shutter closes on image: ", dark_index)
# Set the [start, end] index of the blocked images, flat and dark.
flat_range = [0, 1]
data_range = [48, dark_index]
dark_range = [dark_index, nfile]
# # for fast testing
# data_range = [48, dark_index]
flat = cdata[flat_range[0]:flat_range[1], :, :]
proj = cdata[data_range[0]:data_range[1], :, :]
dark = np.zeros((dark_range[1]-dark_range[0], proj.shape[1], proj.shape[2]))
# if you want to use the shutter closed images as dark uncomment this:
#dark = cdata[dark_range[0]:dark_range[1], :, :]
ndata = tomopy.normalize(proj, flat, dark)
ndata = tomopy.normalize_bg(ndata, air=ndata.shape[2]/2.5)
ndata = tomopy.minus_log(ndata)
sharpening(ndata)
slider(ndata)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=0,
help="index of the first image: 100 (default 0)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
# View the data
slider(rdata)
# Apply the sobel filter
ndata = scale_to_one(rdata)
ndata = sobel_stack(ndata)
slider(ndata)
blur_radius = 3.0
threshold = .04
nddata = label(ndata, blur_radius, threshold)
slider(ndata)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument("top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start", nargs='?', const=1, type=int, default=0, help="index of the first image: 100 (default 0)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[1]
nfile = len(fnmatch.filter(os.listdir(top), '*.tiff'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
# Read the tiff raw data.
rdata = dxchange.read_tiff_stack(fname, ind=ind_tomo)
slider(rdata)
return data
if __name__ == "__main__":
in_file = sys.argv[1]
order = int(sys.argv[2])
iter = int(sys.argv[3])
binning = int(sys.argv[4])
#idF = in_file.find('rect')
out_file = in_file + 'rotated' + str(order) + '_' + str(iter) + '_' + str(
binning)
print('rotate', out_file)
#data = dxchange.read_tiff_stack(in_file+'/results_admm/u/r_00000.tiff', ind=range(0, 2048))
data = dxchange.read_tiff_stack(in_file + '/data/of_recon/recon/iter' +
str(iter) + '_00000.tiff',
ind=range(0, 2048 // pow(2, binning)))
data = rotate_batch(data, 51, order)
data = data.swapaxes(0, 2)
data = rotate_batch(data, 34, order)
data = data.swapaxes(0, 2)
# data = rotate_batch(data, -39)
# data = data.swapaxes(0,1)
# data = rotate_batch(data, 34)
# data = data.swapaxes(0,1)
dxchange.write_tiff_stack(data[400 // pow(2, binning):800 //
pow(2, binning)],
out_file + '/r',
patch_size = (dim_img, dim_img)
batch_size = 50
nb_classes = 2
nb_epoch = 12
# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
nb_pool = 2
# convolution kernel size
nb_conv = 3
fname = '../../test/test_data/1038.tiff'
ind_uncenter1 = range(1038, 1047)
ind_uncenter2 = range(1049, 1057)
uncenter1 = dxchange.read_tiff_stack(fname, ind=ind_uncenter1, digit=4)
uncenter2 = dxchange.read_tiff_stack(fname, ind=ind_uncenter2, digit=4)
uncenter = np.concatenate((uncenter1, uncenter2), axis=0)
uncenter = nor_data(uncenter)
print (uncenter.shape)
uncenter = img_window(uncenter[:, 360:1460, 440:1440], 200)
print (uncenter.shape)
uncenter_patches = extract_3d(uncenter, patch_size, 1)
np.random.shuffle(uncenter_patches)
print (uncenter_patches.shape)
# print uncenter_patches.shape
center_img = dxchange.read_tiff('../../test/test_data/1048.tiff')
center_img = nor_data(center_img)
print (center_img.shape)
center_img = img_window(center_img[360:1460, 440:1440], 400)
center_patches = extract_3d(center_img, patch_size, 1)
radialprofile = (tbinre+1j*tbinim) / np.sqrt(nr)
return radialprofile
if __name__ == "__main__":
fname1 = sys.argv[1]
fname2 = sys.argv[2]
fnameout = sys.argv[3]
wsize = int(sys.argv[4])
pixel = float(sys.argv[5])
frac = float(sys.argv[6])
sslice = int(sys.argv[7])
f1 = dxchange.read_tiff_stack(fname1, ind=np.arange(sslice,sslice+wsize))
f2 = dxchange.read_tiff_stack(fname2, ind=np.arange(sslice,sslice+wsize))
f1 = f1[f1.shape[0]//2-wsize//2:f1.shape[0]//2+wsize // 2, f1.shape[1]//2-wsize //
2:f1.shape[1]//2+wsize//2, f1.shape[2]//2-wsize//2:f1.shape[2]//2+wsize//2]
f2 = f2[f2.shape[0]//2-wsize//2:f2.shape[0]//2+wsize // 2, f2.shape[1]//2-wsize //
2:f2.shape[1]//2+wsize//2, f2.shape[2]//2-wsize//2:f2.shape[2]//2+wsize//2]
print(1)
ff1 = sp.fft.fftshift(sp.fft.fftn(sp.fft.fftshift(f1),workers=-1))
print(ff1.shape)
ff2 = sp.fft.fftshift(sp.fft.fftn(sp.fft.fftshift(f2),workers=-1))
print(3)
frc1 = radial_profile3d(ff1*np.conj(ff2), np.array(ff1.shape)//2) /\
np.sqrt(radial_profile3d(np.abs(ff1)**2, np.array(ff1.shape)//2)
from tomoalign.utils import find_min_max
from skimage.registration import phase_cross_correlation
#from silx.image import sift
from pystackreg import StackReg
data_prefix = '/local/data/vnikitin/nanomax/'
if __name__ == "__main__":
# Model parameters
n = 512 # object size n x,y
nz = 512 # object size in z
nmodes = 4
nscan = 10000
ntheta = 173 # number of angles (rotations)
data = dxchange.read_tiff_stack('sorted/psiangle' + str(nmodes) +
str(nscan) + '/r_00000.tiff',
ind=np.arange(0, 173)).astype('float32')
data_sift = dxchange.read_tiff('siftaligned/data.tif').astype('float32')
print(data.shape)
print(data_sift.shape)
shift = np.zeros((ntheta, 2), dtype='float32')
for k in range(ntheta):
shift0, error, diffphase = phase_cross_correlation(data[k],
data_sift[k],
upsample_factor=1)
print(shift0)
data[k] = np.roll(data[k], (-int(shift0[0]), -int(shift0[1])),
axis=(0, 1))
shift[k] = shift0
dxchange.write_tiff_stack(data,
'sift_aligned_check/psiangle' + str(nmodes) +
patch_size = (dim_img, dim_img)
batch_size = 50
nb_classes = 2
nb_epoch = 12
# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
nb_pool = 2
# convolution kernel size
nb_conv = 3
fname = '../../test/test_data/1038.tiff'
ind_uncenter1 = range(1038, 1047)
ind_uncenter2 = range(1049, 1057)
uncenter1 = dxchange.read_tiff_stack(fname, ind=ind_uncenter1, digit=4)
uncenter2 = dxchange.read_tiff_stack(fname, ind=ind_uncenter2, digit=4)
uncenter = np.concatenate((uncenter1, uncenter2), axis=0)
uncenter = nor_data(uncenter)
print (uncenter.shape)
uncenter = img_window(uncenter[:, 360:1460, 440:1440], 200)
print (uncenter.shape)
uncenter_patches = extract_3d(uncenter, patch_size, 1)
np.random.shuffle(uncenter_patches)
print (uncenter_patches.shape)
# print uncenter_patches.shape
center_img = dxchange.read_tiff('../../test/test_data/1048.tiff')
center_img = nor_data(center_img)
print (center_img.shape)
center_img = img_window(center_img[360:1460, 440:1440], 400)
center_patches = extract_3d(center_img, patch_size, 1)
# for k in range(ndsets):
# data[k*nth:(k+1)*nth] = np.load(fname+'_bin'+str(binning)+str(k)+'.npy').astype('float32')
# theta[k*nth:(k+1)*nth] = np.load(fname+'_theta'+str(k)+'.npy').astype('float32')
# data[np.isnan(data)]=0
# data-=np.mean(data)
# for k in range(7):
# dxchange.write_tiff(data[k*100]-data[0], '/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/proj/d/d_0000'+str(k),overwrite=True)
# dxchange.write_tiff(data[k*100], '/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/proj/r/r_0000'+str(k),overwrite=True)
# exit()
# u = dxchange.read_tiff_stack('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/results_admm/u/r_00000.tiff',ind=np.arange(0,1024))
# ucg = dxchange.read_tiff_stack('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/results_cg/u/r_00000.tiff',ind=np.arange(0,1024))
# ucgn = dxchange.read_tiff_stack('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/revision_cg_nop_1/results/u/r_00000.tiff',ind=np.arange(0,1024))
# un = dxchange.read_tiff_stack('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/results_admm_reg3e-06/u/r_00000.tiff',ind=np.arange(0,1024))
upsi3 = dxchange.read_tiff_stack(
'/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/c2_4_64_80_60_4_FBP_full/tomo_delta__ram-lak_freqscl_1.00_0001.tif',
ind=np.arange(1, 1025))
#upsi7 = dxchange.read_tiff_stack('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/c7_4_64_80_60_4_SART_full/tomo_delta__ram-lak_freqscl_1.00_0001.tif',ind=np.arange(1,1025))
# u= u[:,u.shape[1]//2-612:u.shape[1]//2+612,u.shape[2]//2-612:u.shape[2]//2+612]
# un=un[:,un.shape[1]//2-612:un.shape[1]//2+612,un.shape[2]//2-612:un.shape[2]//2+612]
upsi3 = upsi3[:, upsi3.shape[1] // 2 - 612:upsi3.shape[1] // 2 + 612,
upsi3.shape[2] // 2 - 612:upsi3.shape[2] // 2 + 612]
# vmin=-0.0018
# vmax=0.0018
# a=u[u.shape[0]//2];a[0]=vmin;a[1]=vmax
# plt.imsave('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/figs/uz.png',a,vmin=vmin,vmax=vmax,cmap='gray')
# a=u[:,u.shape[1]//2];a[0]=vmin;a[1]=vmax
# plt.imsave('/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/figs/uy.png',a,vmin=vmin,vmax=vmax,cmap='gray')
# a=u[:,:,u.shape[2]//2];a[0]=vmin;a[1]=vmax
##fname = '/local/decarlo/data/hzg/nanotomography/scan_renamed_450projections_crop_aligned/align_iter_40/radios/image_00000.tiff'
##fname = '/local/decarlo/data/hzg/nanotomography/scan_renamed_450projections_crop_rotate_aligned/align_iter_39/radios/image_00000.tiff'
fname = '/local/decarlo/data/hzg/nanotomography/scan_renamed_450projections_crop_rotate_aligned/align_iter_40/radios/image_00000.tiff'
sample_detector_distance = 18.8e2
detector_pixel_size_x = 19.8e-7
monochromator_energy = 11.0
# for scan_renamed_450projections
proj_start = 0
proj_end = 451
ind_tomo = range(proj_start, proj_end)
# Read normalized, centered and -log() data generated by the Doga's alignment routine.
proj = dxchange.read_tiff_stack(fname, ind=ind_tomo, digit=5)
# Set data collection angles as equally spaced between 0-180 degrees.
theta = tomopy.angles(proj.shape[0])
rot_center = (proj.shape[2]) / 2.0
print("Center of rotation: ", rot_center)
# Reconstruct object using Gridrec algorithm.
rec = tomopy.recon(proj, theta, center=rot_center, algorithm='gridrec')
# Mask each reconstructed slice with a circle.
rec = tomopy.circ_mask(rec, axis=0, ratio=0.95)
# Write data as stack of TIFs.
##fname='/local/decarlo/data/hzg/nanotomography/scan_renamed_450projections_crop_aligned/align_iter_40/recon_dir/aligned_gridrec/recon'
import numpy as np
import dxchange
import sys
import matplotlib.pyplot as plt
name = sys.argv[1]
id = sys.argv[2]
a = dxchange.read_tiff_stack(name + '/r_00000.tiff',
ind=np.arange(0, 512))[:, 64:-64, 64:-64]
print(a.shape)
print(np.linalg.norm(a))
m = np.mean(a[128:-128, 128:-128, 128:-128])
s = np.std(a[128:-128, 128:-128, 128:-128])
a[a > m + 2.5 * s] = m + 2.5 * s
a[a < m - 2.5 * s] = m - 2.5 * s
plt.imsave('figs/z' + str(id) + '.png', a[a.shape[0] // 2], cmap='gray')
plt.imsave('figs/y' + str(id) + '.png', a[:, a.shape[1] // 2], cmap='gray')
plt.imsave('figs/x' + str(id) + '.png', a[:, :, a.shape[2] // 2], cmap='gray')
import os
import glob
import dxchange
import numpy as np
import h5py
print("Working Dir")
print(os.getcwd())
# cube = dxchange.read_tiff_stack('cube/z01.tiff',np.arange(1,25)) #raw data 8bit, change "256" to # of sections
cube1 = dxchange.read_tiff_stack('G0013_cube/z0001.tif', np.arange(
1, 100)) #raw data 8bit, change "256" to # of sections
#cube2 = dxchange.read_tiff_stack('inference_cube_02/z_0000.tif',np.arange(0,75)) #raw data 8bit, change "256" to # of sections
#cube3 = np.append(cube2,cube,axis=0)
# print (cube.shape)
# print ('Mean : '+str(cube.mean()))
# print ('Std : '+str(cube.std()))
print(cube1.shape)
print('Mean : ' + str(cube1.mean()))
print('Std : ' + str(cube1.std()))
#h5file = h5py.File('inference_data.h5', 'w')
#h5file.create_dataset('inference1',data=cube3)
h5file = h5py.File('G0013_cube_data.h5', 'w')
h5file.create_dataset('inference1', data=cube1)
h5file.close()
print("Finished!! Goodbye!!")
import numpy as np
from transform import train_patch, predict_patch, train_filter, predict_filter
batch_size = 2200
nb_epoch = 40
patch_step = 1
nb_filters = 16
nb_conv = 3
patch_size = 64
patch_step = 1
spath = '/home/beams/YANGX/cnn_prj_enhance/tf_prd_battery_20170501/'
ipath = 'weights/tf_mouse.h5'
wpath = 'weights/tf_battery.h5'
proj_start = 1200
proj_end = 1201
ind_tomo = range(proj_start, proj_end)
fname = '/home/beams1/YANGX/cnn_prj_enhance/battery1_ds/prj_00000.tiff'
#
# imgx = dxchange.read_tiff('/home/beams1/YANGX/cnn_prj_enhance/battery1_train/trainx.tif')
# imgy = dxchange.read_tiff('/home/beams1/YANGX/cnn_prj_enhance/battery1_train/trainy.tif')
#
# mdl = train_patch(imgx, imgy, patch_size, 3, nb_filters, nb_conv, batch_size, nb_epoch, ipath)
# mdl.save_weights(wpath)
img_n = dxchange.read_tiff_stack(fname, ind_tomo, digit = 5)
predict_patch(img_n, patch_size, 1, nb_filters, nb_conv, batch_size, wpath, spath)
def main(arg):
parser = argparse.ArgumentParser()
parser.add_argument(
"top", help="top directory where the tiff images are located: /data/")
parser.add_argument("start",
nargs='?',
const=1,
type=int,
default=1,
help="index of the first image: 1000 (default 1)")
args = parser.parse_args()
top = args.top
index_start = int(args.start)
template = os.listdir(top)[0]
nfile = len(fnmatch.filter(os.listdir(top), '*.tif'))
index_end = index_start + nfile
ind_tomo = range(index_start, index_end)
fname = top + template
print(nfile, index_start, index_end, fname)
# Select the sinogram range to reconstruct.
start = 0
end = 512
sino = (start, end)
# Read the tiff raw data.
ndata = dxchange.read_tiff_stack(fname, ind=ind_tomo, slc=(sino, None))
# Normalize to 1 using the air counts
ndata = tomopy.normalize_bg(ndata, air=5)
# Set data collection angles as equally spaced between 0-180 degrees.
theta = tomopy.angles(ndata.shape[0])
ndata = tomopy.minus_log(ndata)
# Set binning and number of iterations
binning = 8
iters = 21
print("Original", ndata.shape)
ndata = tomopy.downsample(ndata, level=binning, axis=1)
# ndata = tomopy.downsample(ndata, level=binning, axis=2)
print("Processing:", ndata.shape)
fdir = 'aligned' + '/noblur_iter_' + str(iters) + '_bin_' + str(binning)
print(fdir)
cprj, sx, sy, conv = alignment.align_seq(ndata,
theta,
fdir=fdir,
iters=iters,
pad=(10, 10),
blur=False,
save=True,
debug=True)
np.save(fdir + '/shift_x', sx)
np.save(fdir + '/shift_y', sy)
# Write aligned projections as stack of TIFs.
dxchange.write_tiff_stack(cprj, fname=fdir + '/radios/image')
tbinre = np.bincount(r.ravel(), data.real.ravel())
tbinim = np.bincount(r.ravel(), data.imag.ravel())
nr = np.bincount(r.ravel())
radialprofile = (tbinre + 1j * tbinim) / np.sqrt(nr)
return radialprofile
wsize = 160
fname1 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm0/u/r_00000.tiff'
fname2 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm1/u/r_00000.tiff'
# fname1 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/brain/Brain_Petrapoxy_day2_2880prj_1440deg_167/cga_resolution1__1440/rect3/r_00000.tiff'
# fname2 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/brain/Brain_Petrapoxy_day2_2880prj_1440deg_167/cga_resolution2__1440/rect3/r_00000.tiff'
f1 = dxchange.read_tiff_stack(fname1, ind=np.arange(0,
320))[:].astype('float32')
f2 = dxchange.read_tiff_stack(fname2, ind=np.arange(0,
320))[:].astype('float32')
f1 = f1[f1.shape[0] // 2, f1.shape[1] // 2 - wsize:f1.shape[1] // 2 + wsize,
f1.shape[2] // 2 - wsize:f1.shape[2] // 2 + wsize]
f2 = f2[f2.shape[0] // 2, f2.shape[1] // 2 - wsize:f2.shape[1] // 2 + wsize,
f2.shape[2] // 2 - wsize:f2.shape[2] // 2 + wsize]
print(f1.shape)
dxchange.write_tiff(
f1, '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/t1')
dxchange.write_tiff(
f2, '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/t2')
ff1 = np.fft.fftshift(np.fft.fft2(f1))
ff2 = np.fft.fftshift(np.fft.fft2(f2))
print(ff1.shape)
print(np.array(ff1.shape) // 2)
import os
import glob
import dxchange
import numpy as np
import h5py
print("Working Dir")
print(os.getcwd())
cube = dxchange.read_tiff_stack('cube02/z01.tif', np.arange(
1, 50)) #raw data 8bit, change "256" to # of sections
labels = dxchange.read_tiff_stack('labels02/l01.tif', np.arange(
1, 50)) #label "ground truth" uint 8 or 32
print('Cube Properties!')
print(cube.dtype)
print(cube.shape)
print('Mean : ' + str(cube.mean()))
print('Std : ' + str(cube.std()))
print('Labels Properties!')
print(labels.dtype)
print(labels.shape)
print('Ids Properties!')
ids = np.unique(labels, return_counts=1)
print(ids)
#raf added here to pad256
# cube = np.pad(cube,((115,116),(0,0),(0,0)),'reflect')
# ndsets = 7
# nth = 100
# data = np.zeros([ndsets*nth,2048//pow(2,binning),2448//pow(2,binning)],dtype='float32')
# theta = np.zeros(ndsets*nth,dtype='float32')
# for k in range(ndsets):
# data[k*nth:(k+1)*nth] = np.load(fname+'_bin'+str(binning)+str(k)+'.npy').astype('float32')
# theta[k*nth:(k+1)*nth] = np.load(fname+'_theta'+str(k)+'.npy').astype('float32')
# data[np.isnan(data)]=0
# data-=np.mean(data)
# for k in range(7):
# dxchange.write_tiff(data[k*100]-data[0], '/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/proj/d/d_0000'+str(k),overwrite=True)
# dxchange.write_tiff(data[k*100], '/data/staff/tomograms/vviknik/tomoalign_vincent_data/mask/Run4_9_1_40min_8keV_phase_100proj_per_rot_interlaced_1201prj_1s_024/proj/r/r_0000'+str(k),overwrite=True)
# exit()
u = dxchange.read_tiff_stack(
'/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm/u/r_00000.tiff',
ind=np.arange(0, 320))[10:-10, 70:-70, 70:-70]
ucg = dxchange.read_tiff_stack(
'/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_cg/u/r_00000.tiff',
ind=np.arange(0, 320))[10:-10, 70:-70, 70:-70]
un = dxchange.read_tiff_stack(
'/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm_reg4e-06/u/r_00000.tiff',
ind=np.arange(0, 320))[10:-10, 70:-70, 70:-70]
ucgn = dxchange.read_tiff_stack(
'/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/TIFF_beta_ram-lak_freqscl_1.00_nonrigid_SART/tomo_delta__ram-lak_freqscl_1.00_0001.tif',
ind=np.arange(1, 321)).astype('float32')[10:-10, 70:-70,
70:-70] #.swapaxes(1,2)
vmax = -5e-4
vmin = -0.006
def reconstruct_fullfield(fname, theta_st=0, theta_end=PI, n_epochs='auto', crit_conv_rate=0.03, max_nepochs=200,
alpha=1e-7, alpha_d=None, alpha_b=None, gamma=1e-6, learning_rate=1.0,
output_folder=None, minibatch_size=None, save_intermediate=False, full_intermediate=False,
energy_ev=5000, psize_cm=1e-7, n_epochs_mask_release=None, cpu_only=False, save_path='.',
phantom_path='phantom', shrink_cycle=20, core_parallelization=True, free_prop_cm=None,
multiscale_level=1, n_epoch_final_pass=None, initial_guess=None, n_batch_per_update=5,
dynamic_rate=True, probe_type='plane', probe_initial=None, probe_learning_rate=1e-3,
pupil_function=None, theta_downsample=None, forward_algorithm='fresnel', random_theta=True,
object_type='normal', **kwargs):
"""
Reconstruct a beyond depth-of-focus object.
:param fname: Filename and path of raw data file. Must be in HDF5 format.
:param theta_st: Starting rotation angle.
:param theta_end: Ending rotation angle.
:param n_epochs: Number of epochs to be executed. If given 'auto', optimizer will stop
when reduction rate of loss function goes below crit_conv_rate.
:param crit_conv_rate: Reduction rate of loss function below which the optimizer should
stop.
:param max_nepochs: The maximum number of epochs to be executed if n_epochs is 'auto'.
:param alpha: Weighting coefficient for both delta and beta regularizer. Should be None
if alpha_d and alpha_b are specified.
:param alpha_d: Weighting coefficient for delta regularizer.
:param alpha_b: Weighting coefficient for beta regularizer.
:param gamma: Weighting coefficient for TV regularizer.
:param learning_rate: Learning rate of ADAM.
:param output_folder: Name of output folder. Put None for auto-generated pattern.
:param downsample: Downsampling (not implemented yet).
:param minibatch_size: Size of minibatch.
:param save_intermediate: Whether to save the object after each epoch.
:param energy_ev: Beam energy in eV.
:param psize_cm: Pixel size in cm.
:param n_epochs_mask_release: The number of epochs after which the finite support mask
is released. Put None to disable this feature.
:param cpu_only: Whether to disable GPU.
:param save_path: The location of finite support mask, the prefix of output_folder and
other metadata.
:param phantom_path: The location of phantom objects (for test version only).
:param shrink_cycle: Shrink-wrap is executed per every this number of epochs.
:param core_parallelization: Whether to use Horovod for parallelized computation within
this function.
:param free_prop_cm: The distance to propagate the wavefront in free space after exiting
the sample, in cm.
:param multiscale_level: The level of multiscale processing. When this number is m and
m > 1, m - 1 low-resolution reconstructions will be performed
before reconstructing with the original resolution. The downsampling
factor for these coarse reconstructions will be [2^(m - 1),
2^(m - 2), ..., 2^1].
:param n_epoch_final_pass: specify a number of iterations for the final pass if multiscale
is activated. If None, it will be the same as n_epoch.
:param initial_guess: supply an initial guess. If None, object will be initialized with noises.
:param n_batch_per_update: number of minibatches during which gradients are accumulated, after
which obj is updated.
:param dynamic_rate: when n_batch_per_update > 1, adjust learning rate dynamically to allow it
to decrease with epoch number
:param probe_type: type of wavefront. Can be 'plane', ' fixed', or 'optimizable'. If 'optimizable',
the probe function will be optimized along with the object.
:param probe_initial: can be provided for 'optimizable' probe_type, and must be provided for
'fixed'.
"""
def rotate_and_project(i, loss, obj_delta, obj_beta):
warnings.warn('Obsolete function. The output loss is scaled by minibatch_size. Proceed with caution.')
obj_rot = tf_rotate(tf.stack([obj_delta, obj_beta], axis=-1), this_theta_batch[i], interpolation='BILINEAR')
if not cpu_only:
with tf.device('/gpu:0'):
exiting = multislice_propagate(obj_rot[:, :, :, 0], obj_rot[:, :, :, 1], probe_real, probe_imag, energy_ev, psize_cm * ds_level, h=h, free_prop_cm=free_prop_cm)
else:
exiting = multislice_propagate(obj_rot[:, :, :, 0], obj_rot[:, :, :, 1], probe_real, probe_imag, energy_ev, psize_cm * ds_level, h=h, free_prop_cm=free_prop_cm)
loss += tf.reduce_mean(tf.squared_difference(tf.abs(exiting), tf.abs(this_prj_batch[i])))
# i = tf.add(i, 1)
return (i, loss, obj)
def rotate_and_project_batch(obj_delta, obj_beta):
obj_rot_batch = []
for i in range(minibatch_size):
obj_rot_batch.append(tf_rotate(tf.stack([obj_delta, obj_beta], axis=-1), this_theta_batch[i], interpolation='BILINEAR'))
# obj_rot = apply_rotation(obj, coord_ls[rand_proj], 'arrsize_64_64_64_ntheta_500')
obj_rot_batch = tf.stack(obj_rot_batch)
if probe_type == 'point':
exiting_batch = multislice_propagate_spherical(obj_rot_batch[:, :, :, :, 0], obj_rot_batch[:, :, :, :, 1],
probe_real, probe_imag, energy_ev,
psize_cm * ds_level, dist_to_source_cm, det_psize_cm,
theta_max, phi_max, free_prop_cm,
obj_batch_shape=[minibatch_size, *obj_size])
else:
if forward_algorithm == 'fresnel':
exiting_batch = multislice_propagate_batch(obj_rot_batch[:, :, :, :, 0], obj_rot_batch[:, :, :, :, 1],
probe_real, probe_imag, energy_ev,
psize_cm * ds_level, free_prop_cm=free_prop_cm, obj_batch_shape=[minibatch_size, *obj_size])
elif forward_algorithm == 'fd':
exiting_batch = multislice_propagate_fd(obj_rot_batch[:, :, :, :, 0], obj_rot_batch[:, :, :, :, 1],
probe_real, probe_imag, energy_ev,
psize_cm * ds_level, free_prop_cm=free_prop_cm,
obj_batch_shape=[minibatch_size, *obj_size])
loss = tf.reduce_mean(tf.squared_difference(tf.abs(exiting_batch), tf.abs(this_prj_batch)), name='loss')
return loss, exiting_batch
# import Horovod or its fake shell
if core_parallelization is False:
warnings.warn('Parallelization is disabled in the reconstruction routine. ')
from pseudo import hvd
else:
try:
import horovod.tensorflow as hvd
hvd.init()
except:
from pseudo import Hvd
hvd = Hvd()
warnings.warn('Unable to import Horovod.')
try:
assert hvd.mpi_threads_supported()
except:
warnings.warn('MPI multithreading is not supported.')
try:
import mpi4py.rc
mpi4py.rc.initialize = False
from mpi4py import MPI
comm = MPI.COMM_WORLD
mpi4py_is_ok = True
assert hvd.size() == comm.Get_size()
except:
warnings.warn('Unable to import mpi4py. Using multiple threads with n_epoch set to "auto" may lead to undefined behaviors.')
from pseudo import Mpi
comm = Mpi()
mpi4py_is_ok = False
# global_step = tf.Variable(0, trainable=False, name='global_step')
t0 = time.time()
# read data
print_flush('Reading data...')
f = h5py.File(os.path.join(save_path, fname), 'r')
prj_0 = f['exchange/data'][...].astype('complex64')
theta = -np.linspace(theta_st, theta_end, prj_0.shape[0], dtype='float32')
n_theta = len(theta)
prj_theta_ind = np.arange(n_theta, dtype=int)
if theta_downsample is not None:
prj_0 = prj_0[::theta_downsample]
theta = theta[::theta_downsample]
prj_theta_ind = prj_theta_ind[::theta_downsample]
n_theta = len(theta)
original_shape = prj_0.shape
comm.Barrier()
print_flush('Data reading: {} s'.format(time.time() - t0))
print_flush('Data shape: {}'.format(original_shape))
comm.Barrier()
if probe_type == 'point':
dist_to_source_cm = kwargs['dist_to_source_cm']
det_psize_cm = kwargs['det_psize_cm']
theta_max = kwargs['theta_max']
phi_max = kwargs['phi_max']
initializer_flag = False
if output_folder is None:
output_folder = 'recon_360_minibatch_{}_' \
'mskrls_{}_' \
'shrink_{}_' \
'iter_{}_' \
'alphad_{}_' \
'alphab_{}_' \
'gamma_{}_' \
'rate_{}_' \
'energy_{}_' \
'size_{}_' \
'ntheta_{}_' \
'prop_{}_' \
'ms_{}_' \
'cpu_{}' \
.format(minibatch_size, n_epochs_mask_release, shrink_cycle,
n_epochs, alpha_d, alpha_b,
gamma, learning_rate, energy_ev,
prj_0.shape[-1], prj_0.shape[0], free_prop_cm,
multiscale_level, cpu_only)
if abs(PI - theta_end) < 1e-3:
output_folder += '_180'
if save_path != '.':
output_folder = os.path.join(save_path, output_folder)
for ds_level in range(multiscale_level - 1, -1, -1):
graph = tf.Graph()
graph.as_default()
ds_level = 2 ** ds_level
print_flush('Multiscale downsampling level: {}'.format(ds_level))
comm.Barrier()
# downsample data
prj = np.copy(prj_0)
if ds_level > 1:
prj = prj[:, ::ds_level, ::ds_level]
prj = prj.astype('complex64')
comm.Barrier()
dim_y, dim_x = prj.shape[-2:]
if random_theta:
prj_dataset = tf.data.Dataset.from_tensor_slices((theta, prj)).shard(hvd.size(), hvd.rank()).shuffle(
buffer_size=100).repeat().batch(minibatch_size)
else:
prj_dataset = tf.data.Dataset.from_tensor_slices((theta, prj)).shard(hvd.size(), hvd.rank()).repeat().batch(minibatch_size)
prj_iter = prj_dataset.make_one_shot_iterator()
this_theta_batch, this_prj_batch = prj_iter.get_next()
comm.Barrier()
# # read rotation data
# try:
# coord_ls = read_all_origin_coords('arrsize_64_64_64_ntheta_500', n_theta)
# except:
# save_rotation_lookup([dim_y, dim_x, dim_x], n_theta)
# coord_ls = read_all_origin_coords('arrsize_64_64_64_ntheta_500', n_theta)
if minibatch_size is None:
minibatch_size = n_theta
if n_epochs_mask_release is None:
n_epochs_mask_release = np.inf
# =============== finite support mask ==============
try:
mask = dxchange.read_tiff_stack(os.path.join(save_path, 'fin_sup_mask', 'mask_00000.tiff'), range(prj_0.shape[1]), 5)
except:
try:
mask = dxchange.read_tiff(os.path.join(save_path, 'fin_sup_mask', 'mask.tiff'))
except:
obj_pr = dxchange.read_tiff_stack(os.path.join(save_path, 'paganin_obj/recon_00000.tiff'), range(prj_0.shape[1]), 5)
obj_pr = gaussian_filter(np.abs(obj_pr), sigma=3, mode='constant')
mask = np.zeros_like(obj_pr)
mask[obj_pr > 1e-5] = 1
dxchange.write_tiff_stack(mask, os.path.join(save_path, 'fin_sup_mask/mask'), dtype='float32', overwrite=True)
if ds_level > 1:
mask = mask[::ds_level, ::ds_level, ::ds_level]
mask_np = mask
mask = tf.convert_to_tensor(mask, dtype=tf.float32, name='mask')
dim_z = mask.shape[-1]
# unify random seed for all threads
comm.Barrier()
seed = int(time.time() / 60)
np.random.seed(seed)
comm.Barrier()
# initializer_flag = True
np.random.seed(int(time.time()))
if initializer_flag == False:
if initial_guess is None:
print_flush('Initializing with Gaussian random.')
# grid_delta = np.load(os.path.join(phantom_path, 'grid_delta.npy'))
# grid_beta = np.load(os.path.join(phantom_path, 'grid_beta.npy'))
obj_delta_init = np.random.normal(size=[dim_y, dim_x, dim_z], loc=8.7e-7, scale=1e-7) * mask_np
obj_beta_init = np.random.normal(size=[dim_y, dim_x, dim_z], loc=5.1e-8, scale=1e-8) * mask_np
obj_delta_init[obj_delta_init < 0] = 0
obj_beta_init[obj_beta_init < 0] = 0
else:
print_flush('Using supplied initial guess.')
sys.stdout.flush()
obj_delta_init = initial_guess[0]
obj_beta_init = initial_guess[1]
else:
print_flush('Initializing with Gaussian random.')
obj_delta_init = dxchange.read_tiff(os.path.join(output_folder, 'delta_ds_{}.tiff'.format(ds_level * 2)))
obj_beta_init = dxchange.read_tiff(os.path.join(output_folder, 'beta_ds_{}.tiff'.format(ds_level * 2)))
obj_delta_init = upsample_2x(obj_delta_init)
obj_beta_init = upsample_2x(obj_beta_init)
obj_delta_init += np.random.normal(size=[dim_y, dim_x, dim_z], loc=8.7e-7, scale=1e-7) * mask_np
obj_beta_init += np.random.normal(size=[dim_y, dim_x, dim_z], loc=5.1e-8, scale=1e-8) * mask_np
obj_delta_init[obj_delta_init < 0] = 0
obj_beta_init[obj_beta_init < 0] = 0
obj_size = obj_delta_init.shape
if object_type == 'phase_only':
obj_beta_init[...] = 0
obj_delta = tf.Variable(initial_value=obj_delta_init, dtype=tf.float32, name='obj_delta')
obj_beta = tf.constant(obj_beta_init, dtype=tf.float32, name='obj_beta')
elif object_type == 'absorption_only':
obj_delta_init[...] = 0
obj_delta = tf.constant(obj_delta_init, dtype=tf.float32, name='obj_delta')
obj_beta = tf.Variable(initial_value=obj_beta_init, dtype=tf.float32, name='obj_beta')
else:
obj_delta = tf.Variable(initial_value=obj_delta_init, dtype=tf.float32, name='obj_delta')
obj_beta = tf.Variable(initial_value=obj_beta_init, dtype=tf.float32, name='obj_beta')
# ====================================================
if probe_type == 'plane':
probe_real = tf.constant(np.ones([dim_y, dim_x]), dtype=tf.float32)
probe_imag = tf.constant(np.zeros([dim_y, dim_x]), dtype=tf.float32)
elif probe_type == 'optimizable':
if probe_initial is not None:
probe_mag, probe_phase = probe_initial
probe_real, probe_imag = mag_phase_to_real_imag(probe_mag, probe_phase)
else:
# probe_mag = np.ones([dim_y, dim_x])
# probe_phase = np.zeros([dim_y, dim_x])
back_prop_cm = (free_prop_cm + (psize_cm * obj_size[2])) if free_prop_cm is not None else (psize_cm * obj_size[2])
probe_init = create_probe_initial_guess(os.path.join(save_path, fname), back_prop_cm * 1.e7, energy_ev, psize_cm * 1.e7)
probe_real = probe_init.real
probe_imag = probe_init.imag
if pupil_function is not None:
probe_real = probe_real * pupil_function
probe_imag = probe_imag * pupil_function
pupil_function = tf.convert_to_tensor(pupil_function, dtype=tf.float32)
probe_real = tf.Variable(probe_real, dtype=tf.float32, trainable=True)
probe_imag = tf.Variable(probe_imag, dtype=tf.float32, trainable=True)
elif probe_type == 'fixed':
probe_mag, probe_phase = probe_initial
probe_real, probe_imag = mag_phase_to_real_imag(probe_mag, probe_phase)
probe_real = tf.constant(probe_real, dtype=tf.float32)
probe_imag = tf.constant(probe_imag, dtype=tf.float32)
elif probe_type == 'point':
# this should be in spherical coordinates
probe_real = tf.constant(np.ones([dim_y, dim_x]), dtype=tf.float32)
probe_imag = tf.constant(np.zeros([dim_y, dim_x]), dtype=tf.float32)
elif probe_type == 'gaussian':
probe_mag_sigma = kwargs['probe_mag_sigma']
probe_phase_sigma = kwargs['probe_phase_sigma']
probe_phase_max = kwargs['probe_phase_max']
py = np.arange(obj_size[0]) - (obj_size[0] - 1.) / 2
px = np.arange(obj_size[1]) - (obj_size[1] - 1.) / 2
pxx, pyy = np.meshgrid(px, py)
probe_mag = np.exp(-(pxx ** 2 + pyy ** 2) / (2 * probe_mag_sigma ** 2))
probe_phase = probe_phase_max * np.exp(
-(pxx ** 2 + pyy ** 2) / (2 * probe_phase_sigma ** 2))
probe_real, probe_imag = mag_phase_to_real_imag(probe_mag, probe_phase)
probe_real = tf.constant(probe_real, dtype=tf.float32)
probe_imag = tf.constant(probe_imag, dtype=tf.float32)
else:
raise ValueError('Invalid wavefront type. Choose from \'plane\', \'fixed\', \'optimizable\'.')
# =============finite support===================
obj_delta = obj_delta * mask
obj_beta = obj_beta * mask
# obj_delta = tf.nn.relu(obj_delta)
# obj_beta = tf.nn.relu(obj_beta)
# ==============================================
# ================shrink wrap===================
def shrink_wrap():
boolean = tf.cast(obj_delta > 1e-15, dtype=tf.float32)
_mask = mask * boolean
return _mask
def return_mask(): return mask
# if initializer_flag == False:
i_epoch = tf.Variable(0, trainable=False, dtype='float32')
if shrink_cycle is not None:
mask = tf.cond(tf.greater(i_epoch, shrink_cycle), shrink_wrap, return_mask)
# if hvd.rank() == 0 and hvd.local_rank() == 0:
# dxchange.write_tiff(np.squeeze(sess.run(mask)),
# os.path.join(save_path, 'fin_sup_mask/runtime_mask/epoch_{}'.format(epoch)), dtype='float32', overwrite=True)
# ==============================================
# generate Fresnel kernel
voxel_nm = np.array([psize_cm] * 3) * 1.e7 * ds_level
lmbda_nm = 1240. / energy_ev
delta_nm = voxel_nm[-1]
kernel = get_kernel(delta_nm, lmbda_nm, voxel_nm, [dim_y, dim_y, dim_x])
h = tf.convert_to_tensor(kernel, dtype=tf.complex64, name='kernel')
# loss = tf.constant(0.)
# i = tf.constant(0)
# for j in range(minibatch_size):
# i, loss, obj = rotate_and_project(i, loss, obj)
loss, exiting = rotate_and_project_batch(obj_delta, obj_beta)
# loss = loss / n_theta + alpha * tf.reduce_sum(tf.image.total_variation(obj))
# loss = loss / n_theta + gamma * energy_leak(obj, mask_add)
if alpha_d is None:
reg_term = alpha * (tf.norm(obj_delta, ord=1) + tf.norm(obj_delta, ord=1)) + gamma * total_variation_3d(obj_delta)
else:
if gamma == 0:
reg_term = alpha_d * tf.norm(obj_delta, ord=1) + alpha_b * tf.norm(obj_beta, ord=1)
else:
reg_term = alpha_d * tf.norm(obj_delta, ord=1) + alpha_b * tf.norm(obj_beta, ord=1) + gamma * total_variation_3d(obj_delta)
loss = loss + reg_term
if probe_type == 'optimizable':
probe_reg = 1.e-10 * (tf.image.total_variation(tf.reshape(probe_real, [dim_y, dim_x, -1])) +
tf.image.total_variation(tf.reshape(probe_real, [dim_y, dim_x, -1])))
loss = loss + probe_reg
tf.summary.scalar('loss', loss)
tf.summary.scalar('regularizer', reg_term)
tf.summary.scalar('error', loss - reg_term)
if dynamic_rate and n_batch_per_update > 1:
# modifier = 1. / n_batch_per_update
modifier = tf.exp(-i_epoch) * (n_batch_per_update - 1) + 1
optimizer = tf.train.AdamOptimizer(learning_rate=float(learning_rate) * hvd.size() * modifier)
else:
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate * hvd.size())
optimizer = hvd.DistributedOptimizer(optimizer, name='distopt_{}'.format(ds_level))
if n_batch_per_update > 1:
accum_grad_delta = tf.Variable(tf.zeros_like(obj_delta.initialized_value()), trainable=False)
accum_grad_beta = tf.Variable(tf.zeros_like(obj_beta.initialized_value()), trainable=False)
this_grad_delta = optimizer.compute_gradients(loss, obj_delta)
this_grad_delta = this_grad_delta[0]
this_grad_beta = optimizer.compute_gradients(loss, this_grad_beta)
this_grad_beta = this_grad_beta[0]
initialize_grad_delta = accum_grad_delta.assign(tf.zeros_like(accum_grad_delta))
initialize_grad_beta = accum_grad_beta.assign(tf.zeros_like(accum_grad_beta))
accum__op_delta = accum_grad_delta.assign_add(this_grad_delta[0])
accum_op_beta = accum_grad_beta.assign_add(this_grad_beta[0])
update_obj_delta = optimizer.apply_gradients([(accum_grad_delta / n_batch_per_update, this_grad_beta[1])])
update_obj_beta = optimizer.apply_gradients([(accum_grad_beta / n_batch_per_update, this_grad_beta[1])])
else:
###
this_grad_delta = optimizer.compute_gradients(loss, obj_delta)
###
if object_type == 'normal':
optimizer = optimizer.minimize(loss)
elif object_type == 'phase_only':
optimizer = optimizer.minimize(loss)
elif object_type == 'absorption_only':
optimizer = optimizer.minimize(loss)
else:
raise ValueError
# if minibatch_size >= n_theta:
# optimizer = optimizer.minimize(loss, var_list=[obj])
# hooks = [hvd.BroadcastGlobalVariablesHook(0)]
if probe_type == 'optimizable':
optimizer_probe = tf.train.AdamOptimizer(learning_rate=probe_learning_rate * hvd.size())
optimizer_probe = hvd.DistributedOptimizer(optimizer_probe, name='distopt_probe_{}'.format(ds_level))
if n_batch_per_update > 1:
accum_grad_probe = [tf.Variable(tf.zeros_like(probe_real.initialized_value()), trainable=False),
tf.Variable(tf.zeros_like(probe_imag.initialized_value()), trainable=False)]
this_grad_probe = optimizer_probe.compute_gradients(loss, [probe_real, probe_imag])
initialize_grad_probe = [accum_grad_probe[i].assign(tf.zeros_like(accum_grad_probe[i])) for i in range(2)]
accum_op_probe = [accum_grad_probe[i].assign_add(this_grad_probe[i][0]) for i in range(2)]
update_probe = [optimizer_probe.apply_gradients([(accum_grad_probe[i] / n_batch_per_update, this_grad_probe[i][1])]) for i in range(2)]
else:
optimizer_probe = optimizer_probe.minimize(loss, var_list=[probe_real, probe_imag])
if minibatch_size >= n_theta:
optimizer_probe = optimizer_probe.minimize(loss, var_list=[probe_real, probe_imag])
# =============finite support===================
obj_delta = obj_delta * mask
obj_beta = obj_beta * mask
obj_delta = tf.nn.relu(obj_delta)
obj_beta = tf.nn.relu(obj_beta)
# ==============================================
loss_ls = []
reg_ls = []
merged_summary_op = tf.summary.merge_all()
# create benchmarking metadata
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
if cpu_only:
sess = tf.Session(config=tf.ConfigProto(device_count = {'GPU': 0}, allow_soft_placement=True))
else:
config = tf.ConfigProto(log_device_placement=False)
config.gpu_options.allow_growth = True
config.gpu_options.visible_device_list = str(hvd.local_rank())
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
hvd.broadcast_global_variables(0)
if hvd.rank() == 0:
preset = 'pp' if probe_type == 'point' else 'fullfield'
create_summary(output_folder, locals(), preset=preset)
summary_writer = tf.summary.FileWriter(os.path.join(output_folder, 'tb'), sess.graph)
t0 = time.time()
print_flush('Optimizer started.')
n_loop = n_epochs if n_epochs != 'auto' else max_nepochs
if ds_level == 1 and n_epoch_final_pass is not None:
n_loop = n_epoch_final_pass
n_batch = int(np.ceil(float(n_theta) / minibatch_size) / hvd.size())
t00 = time.time()
for epoch in range(n_loop):
if mpi4py_is_ok:
stop_iteration = False
else:
stop_iteration_file = open('.stop_itertion', 'w')
stop_iteration_file.write('False')
stop_iteration_file.close()
i_epoch = i_epoch + 1
if minibatch_size <= n_theta:
batch_counter = 0
for i_batch in range(n_batch):
try:
if n_batch_per_update > 1:
t0_batch = time.time()
if probe_type == 'optimizable':
_, _, _, current_loss, current_reg, current_probe_reg, summary_str = sess.run(
[accum__op_delta, accum_op_beta, accum_op_probe, loss, reg_term, probe_reg, merged_summary_op], options=run_options,
run_metadata=run_metadata)
else:
_, _, current_loss, current_reg, summary_str = sess.run(
[accum__op_delta, accum_op_beta, loss, reg_term, merged_summary_op], options=run_options,
run_metadata=run_metadata)
print_flush('Minibatch done in {} s (rank {}); current loss = {}, probe reg. = {}.'.format(time.time() - t0_batch, hvd.rank(), current_loss, current_probe_reg))
batch_counter += 1
if batch_counter == n_batch_per_update or i_batch == n_batch - 1:
sess.run([update_obj_delta, update_obj_beta])
sess.run([initialize_grad_delta, initialize_grad_beta])
if probe_type == 'optimizable':
sess.run(update_probe)
sess.run(initialize_grad_probe)
batch_counter = 0
print_flush('Gradient applied.')
else:
t0_batch = time.time()
if probe_type == 'optimizable':
_, _, current_loss, current_reg, current_probe_reg, summary_str = sess.run([optimizer, optimizer_probe, loss, reg_term, probe_reg, merged_summary_op], options=run_options, run_metadata=run_metadata)
print_flush(
'Minibatch done in {} s (rank {}); current loss = {}, probe reg. = {}.'.format(
time.time() - t0_batch, hvd.rank(), current_loss, current_probe_reg))
else:
_, current_loss, current_reg, summary_str, mask_int, current_obj, current_grad = sess.run([optimizer, loss, reg_term, merged_summary_op, mask, obj_delta, this_grad_delta], options=run_options, run_metadata=run_metadata)
print_flush(
'Minibatch done in {} s (rank {}); current loss = {}; current reg = {}.'.format(
time.time() - t0_batch, hvd.rank(), current_loss, current_reg))
# dxchange.write_tiff(current_obj[int(105./256*current_obj.shape[0])], os.path.join(output_folder, 'intermediate_minibatch/delta'), dtype='float32')
# dxchange.write_tiff(current_grad[0][0][int(105./256*current_obj.shape[0])], os.path.join(output_folder, 'intermediate_minibatch/grad'), dtype='float32')
# dxchange.write_tiff(abs(mask_int), os.path.join(output_folder, 'masks', 'mask_{}'.format(epoch)), dtype='float32')
# dxchange.write_tiff(exiting_wave, save_path + '/exit', dtype='float32')
# enforce pupil function
if probe_type == 'optimizable' and pupil_function is not None:
probe_real = probe_real * pupil_function
probe_imag = probe_imag * pupil_function
except tf.errors.OutOfRangeError:
break
else:
if probe_type == 'optimizable':
_, _, current_loss, current_reg, summary_str = sess.run([optimizer, optimizer_probe, loss, reg_term, merged_summary_op], options=run_options, run_metadata=run_metadata)
else:
_, current_loss, current_reg, summary_str = sess.run([optimizer, loss, reg_term, merged_summary_op], options=run_options, run_metadata=run_metadata)
# timeline for benchmarking
if hvd.rank() == 0:
tl = timeline.Timeline(run_metadata.step_stats)
ctf = tl.generate_chrome_trace_format()
try:
os.makedirs(os.path.join(output_folder, 'profiling'))
except:
pass
with open(os.path.join(output_folder, 'profiling', 'time_{}.json'.format(epoch)), 'w') as f:
f.write(ctf)
f.close()
# check stopping criterion
if n_epochs == 'auto':
if len(loss_ls) > 0:
print_flush('Reduction rate of loss is {}.'.format((current_loss - loss_ls[-1]) / loss_ls[-1]))
sys.stdout.flush()
if len(loss_ls) > 0 and -crit_conv_rate < (current_loss - loss_ls[-1]) / loss_ls[-1] < 0 and hvd.rank() == 0:
loss_ls.append(current_loss)
reg_ls.append(current_reg)
summary_writer.add_summary(summary_str, epoch)
if mpi4py_is_ok:
stop_iteration = True
else:
stop_iteration = open('.stop_iteration', 'w')
stop_iteration.write('True')
stop_iteration.close()
comm.Barrier()
if mpi4py_is_ok:
stop_iteration = comm.bcast(stop_iteration, root=0)
else:
stop_iteration_file = open('.stop_iteration', 'r')
stop_iteration = stop_iteration_file.read()
stop_iteration_file.close()
stop_iteration = True if stop_iteration == 'True' else False
if stop_iteration:
break
# if epoch < n_epochs_mask_release:
# # ================shrink wrap===================
# if epoch % shrink_cycle == 0 and epoch > 0:
# mask_temp = sess.run(obj_delta > 1e-8)
# boolean = tf.convert_to_tensor(mask_temp, dtype=tf.float32)
# mask = mask * boolean
# if hvd.rank() == 0 and hvd.local_rank() == 0:
# dxchange.write_tiff(np.squeeze(sess.run(mask)),
# os.path.join(save_path, 'fin_sup_mask/runtime_mask/epoch_{}'.format(epoch)), dtype='float32', overwrite=True)
# # ==============================================
# =============finite support===================
# obj_delta = obj_delta * mask
# obj_beta = obj_beta * mask
# obj_delta = tf.nn.relu(obj_delta)
# obj_beta = tf.nn.relu(obj_beta)
# ==============================================
if hvd.rank() == 0:
loss_ls.append(current_loss)
reg_ls.append(current_reg)
summary_writer.add_summary(summary_str, epoch)
if save_intermediate and hvd.rank() == 0:
temp_obj_delta, temp_obj_beta = sess.run([obj_delta, obj_beta])
temp_obj_delta = np.abs(temp_obj_delta)
temp_obj_beta = np.abs(temp_obj_beta)
if full_intermediate:
dxchange.write_tiff(temp_obj_delta,
fname=os.path.join(output_folder, 'intermediate', 'ds_{}_iter_{:03d}'.format(ds_level, epoch)),
dtype='float32',
overwrite=True)
else:
dxchange.write_tiff(temp_obj_delta[int(temp_obj_delta.shape[0] / 2), :, :],
fname=os.path.join(output_folder, 'intermediate', 'ds_{}_iter_{:03d}'.format(ds_level, epoch)),
dtype='float32',
overwrite=True)
probe_current_real, probe_current_imag = sess.run([probe_real, probe_imag])
probe_current_mag, probe_current_phase = real_imag_to_mag_phase(probe_current_real, probe_current_imag)
dxchange.write_tiff(probe_current_mag,
fname=os.path.join(output_folder, 'intermediate', 'probe',
'mag_ds_{}_iter_{:03d}'.format(ds_level, epoch)),
dtype='float32',
overwrite=True)
dxchange.write_tiff(probe_current_phase,
fname=os.path.join(output_folder, 'intermediate', 'probe',
'phase_ds_{}_iter_{:03d}'.format(ds_level, epoch)),
dtype='float32',
overwrite=True)
dxchange.write_tiff(temp_obj_delta, os.path.join(output_folder, 'current', 'delta'), dtype='float32', overwrite=True)
print_flush('Iteration {} (rank {}); loss = {}; time = {} s'.format(epoch, hvd.rank(), current_loss, time.time() - t00))
sys.stdout.flush()
# except:
# # if one thread breaks out after meeting stopping criterion, intercept Horovod error and break others
# break
print_flush('Total time: {}'.format(time.time() - t0))
sys.stdout.flush()
if hvd.rank() == 0:
res_delta, res_beta = sess.run([obj_delta, obj_beta])
res_delta *= mask_np
res_beta *= mask_np
res_delta = np.clip(res_delta, 0, None)
res_beta = np.clip(res_beta, 0, None)
dxchange.write_tiff(res_delta, fname=os.path.join(output_folder, 'delta_ds_{}'.format(ds_level)), dtype='float32', overwrite=True)
dxchange.write_tiff(res_beta, fname=os.path.join(output_folder, 'beta_ds_{}'.format(ds_level)), dtype='float32', overwrite=True)
probe_final_real, probe_final_imag = sess.run([probe_real, probe_imag])
probe_final_mag, probe_final_phase = real_imag_to_mag_phase(probe_final_real, probe_final_imag)
dxchange.write_tiff(probe_final_mag, fname=os.path.join(output_folder, 'probe_mag_ds_{}'.format(ds_level)), dtype='float32', overwrite=True)
dxchange.write_tiff(probe_final_phase, fname=os.path.join(output_folder, 'probe_phase_ds_{}'.format(ds_level)), dtype='float32', overwrite=True)
error_ls = np.array(loss_ls) - np.array(reg_ls)
x = len(loss_ls)
plt.figure()
plt.semilogy(range(x), loss_ls, label='Total loss')
plt.semilogy(range(x), reg_ls, label='Regularizer')
plt.semilogy(range(x), error_ls, label='Error term')
plt.legend()
try:
os.makedirs(os.path.join(output_folder, 'convergence'))
except:
pass
plt.savefig(os.path.join(output_folder, 'convergence', 'converge_ds_{}.png'.format(ds_level)), format='png')
np.save(os.path.join(output_folder, 'convergence', 'total_loss_ds_{}'.format(ds_level)), loss_ls)
np.save(os.path.join(output_folder, 'convergence', 'reg_ds_{}'.format(ds_level)), reg_ls)
np.save(os.path.join(output_folder, 'convergence', 'error_ds_{}'.format(ds_level)), error_ls)
print_flush('Clearing current graph...')
sess.run(tf.global_variables_initializer())
sess.close()
tf.reset_default_graph()
initializer_flag = True
print_flush('Current iteration finished.')
'220': 1400,
'221': 3000,
'222': 2200,
}
######################################################
file_name = sys.argv[1]
center = centers[file_name[-6:-3]]
ntheta = nthetas[file_name[-6:-3]]
ngpus = 8
pnz = 16 # chunk size for slices
ptheta = 20 # chunk size for angles
# read data
prj = dxchange.read_tiff_stack(f'{file_name[:-3]}/data/d_00000.tiff',
ind=range(ntheta))
theta = np.load(file_name[:-3] + '/data/theta.npy')
nz, n = prj.shape[1:]
niteradmm = [96, 48, 24, 12] # number of iterations in the ADMM scheme
# niteradmm = [2,2,2] # number of iterations in the ADMM scheme
startwin = [256, 128, 64,
32] # starting window size in optical flow estimation
stepwin = [2, 2, 2,
2] # step for decreasing the window size in optical flow estimtion
res = tomoalign.admm_of_levels(prj,
theta,
pnz,
ptheta,
center,
import dxchange
import numpy as np
import dxchange
from scipy.ndimage import rotate
import sys
import matplotlib.pyplot as plt
if __name__ == "__main__":
in_file = sys.argv[1]
out_file = sys.argv[2]
#mul=2
binning=0
data = dxchange.read_tiff_stack(in_file+'/r_00000.tiff', ind=[691,657,672])
#bin0 cutes
vmin = -0.001*2/3
vmax = 0.002*2/3
plt.imsave(out_file+'2000bin0p1.png',data[0],vmin=vmin,vmax=vmax,cmap='gray')
plt.imsave(out_file+'2000bin0p2.png',data[1],vmin=vmin,vmax=vmax,cmap='gray')
plt.imsave(out_file+'2000bin0p3.png',data[2],vmin=vmin,vmax=vmax,cmap='gray')
tbinre = np.bincount(r.ravel(), data.real.ravel())
tbinim = np.bincount(r.ravel(), data.imag.ravel())
nr = np.bincount(r.ravel())
radialprofile = (tbinre + 1j * tbinim) / np.sqrt(nr)
return radialprofile
wsize = 272
fname1 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm0/u/r_00000.tiff'
fname2 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/psi/d/results_admm1/u/r_00000.tiff'
# fname1 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/brain/Brain_Petrapoxy_day2_2880prj_1440deg_167/cga_resolution1__1440/rect3/r_00000.tiff'
# fname2 = '/data/staff/tomograms/vviknik/tomoalign_vincent_data/brain/Brain_Petrapoxy_day2_2880prj_1440deg_167/cga_resolution2__1440/rect3/r_00000.tiff'
f1 = dxchange.read_tiff_stack(fname1, ind=np.arange(0, 544))[:]
f2 = dxchange.read_tiff_stack(fname2, ind=np.arange(0, 544))[:]
f1 = f1[f1.shape[0] // 2 - wsize:f1.shape[0] // 2 + wsize,
f1.shape[1] // 2 - wsize:f1.shape[1] // 2 + wsize,
f1.shape[2] // 2 - wsize:f1.shape[2] // 2 + wsize]
f2 = f2[f2.shape[0] // 2 - wsize:f2.shape[0] // 2 + wsize,
f2.shape[1] // 2 - wsize:f2.shape[1] // 2 + wsize,
f2.shape[2] // 2 - wsize:f2.shape[2] // 2 + wsize]
print(f1.shape)
ff1 = np.fft.fftshift(np.fft.fftn(np.fft.fftshift(f1)))
ff2 = np.fft.fftshift(np.fft.fftn(np.fft.fftshift(f2)))
frc1 = radial_profile3d(ff1*np.conj(ff2),np.array(ff1.shape)//2)/\
np.sqrt(radial_profile3d(np.abs(ff1)**2,np.array(ff1.shape)//2)*radial_profile3d(np.abs(ff2)**2,np.array(ff1.shape)//2))
# print(np.min(frc1[:wsize].real))