def _process(lock, int_from, int_to, infile, outfile, outshape, outtype, method, plan, logfilename): # Process the required subset of images: for i in range(int_from, int_to + 1): # Read input image: t0 = time() f_in = getHDF5(infile, 'r') if "/tomo" in f_in: dset = f_in['tomo'] else: dset = f_in['exchange/data'] im = tdf.read_tomo(dset,i).astype(float32) f_in.close() t1 = time() # Perform phase retrieval (first time also PyFFTW prepares a plan): if (method == 0): im = tiehom(im, plan).astype(float32) else: im = phrt(im, plan, method).astype(float32) t2 = time() # Save processed image to HDF5 file (atomic procedure - lock used): _write_data(lock, im, i, outfile, outshape, outtype, logfilename, t2 - t1, t1 - t0)
def _process(lock, int_from, int_to, infile, outfile, outshape, outtype,
method, plan, logfilename):
# Process the required subset of images:
for i in range(int_from, int_to + 1):
# Read input image:
t0 = time()
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
im = tdf.read_tomo(dset, i).astype(float32)
f_in.close()
t1 = time()
# Perform phase retrieval (first time also PyFFTW prepares a plan):
if (method == 0):
im = tiehom(im, plan).astype(float32)
elif (method == 1):
im = tiehom2020(im, plan).astype(float32)
else:
im = phrt(im, plan, method).astype(float32)
t2 = time()
# Save processed image to HDF5 file (atomic procedure - lock used):
_write_data(lock, im, i, outfile, outshape, outtype, logfilename,
t2 - t1, t1 - t0)
def process(sino_idx, num_sinos, infile, outpath, preprocessing_required,
corr_plan, norm_sx, norm_dx, flat_end, half_half, half_half_line,
ext_fov, ext_fov_rot_right, ext_fov_overlap, ext_fov_normalize,
ext_fov_average, ringrem, phaseretrieval_required, phrtmethod,
phrt_param1, phrt_param2, energy, distance, pixsize, phrtpad,
approx_win, angles, angles_projfrom, angles_projto, offset,
logtransform, param1, circle, scale, pad, method, zerone_mode,
dset_min, dset_max, decim_factor, downsc_factor, corr_offset,
postprocess_required, convert_opt, crop_opt, nr_threads, off_from,
off_to, logfilename, lock, slice_prefix):
"""To do...
"""
slice_nr = sino_idx
# Perform reconstruction (on-the-fly preprocessing and phase retrieval, if required):
if (phaseretrieval_required):
# In this case a bunch of sinograms is loaded into memory:
#
# Load the temporary data structure reading the input TDF file.
# To know the right dimension the first sinogram is pre-processed.
#
# Open the TDF file and get the dataset:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
# Downscaling and decimation factors considered when determining the approximation window:
zrange = arange(sino_idx - approx_win * downsc_factor / 2,
sino_idx + approx_win * downsc_factor / 2,
downsc_factor)
zrange = zrange[(zrange >= 0)]
zrange = zrange[(zrange < num_sinos)]
approx_win = zrange.shape[0]
# Approximation window cannot be odd:
if (approx_win % 2 == 1):
approx_win = approx_win - 1
zrange = zrange[0:approx_win]
# Read one sinogram to get the proper dimensions:
test_im = tdf.read_sino(dset,
zrange[0] * downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing of the first sinogram to get the right dimension:
if (preprocessing_required):
test_im = flat_fielding(test_im, zrange[0], corr_plan, flat_end,
half_half, half_half_line / decim_factor,
norm_sx, norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(test_im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor,
ext_fov_normalize,
ext_fov_average).astype(float32)
test_im = ring_correction(test_im, ringrem, flat_end,
corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
# Now we can allocate memory for the bunch of slices:
tmp_im = empty((approx_win, test_im.shape[0], test_im.shape[1]),
dtype=float32)
tmp_im[0, :, :] = test_im
# Reading all the the sinos from TDF file and close:
for ct in range(1, approx_win):
test_im = tdf.read_sino(dset,
zrange[ct] * downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing for each sinogram of the bunch:
if (preprocessing_required):
test_im = flat_fielding(test_im, zrange[ct], corr_plan,
flat_end, half_half,
half_half_line / decim_factor, norm_sx,
norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(
test_im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor, ext_fov_normalize,
ext_fov_average).astype(float32)
test_im = ring_correction(test_im, ringrem, flat_end,
corr_plan['skip_flat_after'],
half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
tmp_im[ct, :, :] = test_im
f_in.close()
# Now everything has to refer to a downscaled dataset:
sino_idx = ((zrange == sino_idx).nonzero())
#
# Perform phase retrieval:
#
# Prepare the plan:
if (phrtmethod == 0):
# Paganin's:
phrtplan = tiehom_plan(tmp_im[:, 0, :],
phrt_param1,
phrt_param2,
energy,
distance,
pixsize * downsc_factor,
padding=phrtpad)
else:
phrtplan = phrt_plan(tmp_im[:, 0, :],
energy,
distance,
pixsize * downsc_factor,
phrt_param2,
phrt_param1,
phrtmethod,
padding=phrtpad)
# Process each projection (whose height depends on the size of the bunch):
for ct in range(0, tmp_im.shape[1]):
if (phrtmethod == 0):
tmp_im[:, ct, :] = tiehom(tmp_im[:, ct, :],
phrtplan).astype(float32)
else:
tmp_im[:, ct, :] = phrt(tmp_im[:, ct, :], phrtplan,
phrtmethod).astype(float32)
# Extract the requested sinogram:
im = tmp_im[sino_idx[0], :, :].squeeze()
else:
# Read only one sinogram:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
im = tdf.read_sino(dset, sino_idx * downsc_factor).astype(float32)
f_in.close()
# Apply projection removal (if required):
im = im[angles_projfrom:angles_projto, :]
# Downscale and decimate the sinogram:
im = im[::decim_factor, ::downsc_factor]
#sino_idx = sino_idx/downsc_factor
# Perform the preprocessing of the sinogram (if required):
if (preprocessing_required):
im = flat_fielding(im, sino_idx, corr_plan, flat_end, half_half,
half_half_line / decim_factor, norm_sx,
norm_dx).astype(float32)
if ext_fov:
im = extfov_correction(im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor,
ext_fov_normalize, ext_fov_average)
im = ring_correction(im, ringrem, flat_end,
corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor, ext_fov)
# Log infos:
log = open(logfilename, "a")
log.write(
linesep +
"\tPerforming reconstruction with multiple centers of rotation...")
log.write(linesep + "\t--------------")
log.close()
# Split the computation into multiple processes:
for num in range(nr_threads):
start = ((off_to - off_from + 1) / nr_threads) * num + off_from
if (num == nr_threads - 1):
end = off_to
else:
end = ((off_to - off_from + 1) / nr_threads) * (num +
1) + off_from - 1
Process(target=reconstruct,
args=(im, angles, angles_projfrom, angles_projto,
offset / downsc_factor, logtransform, param1, circle,
scale, pad, method, zerone_mode, dset_min, dset_max,
corr_offset, postprocess_required, convert_opt, crop_opt,
start, end, outpath, slice_nr, downsc_factor,
logfilename, lock, slice_prefix)).start()
def process(sino_idx, num_sinos, infile, outfile, preprocessing_required, corr_plan, skipflat, norm_sx, norm_dx, flat_end, half_half,
half_half_line, ext_fov, ext_fov_rot_right, ext_fov_overlap, ext_fov_normalize, ext_fov_average,
ringrem, phaseretrieval_required, phrtmethod, phrt_param1,
phrt_param2, energy, distance, pixsize, phrtpad, approx_win, angles, angles_projfrom, angles_projto,
offset, logtransform, recpar, circle, scale, pad, method, rolling, roll_shift,
zerone_mode, dset_min, dset_max, decim_factor, downsc_factor, corr_offset, postprocess_required,
polarfilt_opt, convert_opt, crop_opt, dynamic_ff, EFF, filtEFF, im_dark, nr_threads, logfilename, tmppath):
"""To do...
"""
# Perform reconstruction (on-the-fly preprocessing and phase retrieval, if
# required):
if (phaseretrieval_required):
# In this case a bunch of sinograms is loaded into memory:
#
# Load the temporary data structure reading the input TDF file.
# To know the right dimension the first sinogram is pre-processed.
#
# Open the TDF file and get the dataset:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
# Downscaling and decimation factors considered when determining the
# approximation window:
zrange = arange(sino_idx - approx_win * downsc_factor / 2, sino_idx + approx_win * downsc_factor / 2, downsc_factor)
zrange = zrange[(zrange >= 0)]
zrange = zrange[(zrange < num_sinos)]
approx_win = zrange.shape[0]
# Approximation window cannot be odd:
if (approx_win % 2 == 1):
approx_win = approx_win - 1
zrange = zrange[0:approx_win]
# Read one sinogram to get the proper dimensions:
test_im = tdf.read_sino(dset, zrange[0]*downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing of the first sinogram to get the right
# dimension:
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
#test_im = dynamic_flat_fielding(test_im, zrange[0] / downsc_factor, EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
test_im = dynamic_flat_fielding(test_im, zrange[0] , EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
else:
#test_im = flat_fielding(test_im, zrange[0] / downsc_factor, corr_plan, flat_end, half_half,
# half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
test_im = flat_fielding(test_im, zrange[0], corr_plan, flat_end, half_half,
half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(test_im, ext_fov_rot_right, ext_fov_overlap / downsc_factor, ext_fov_normalize, ext_fov_average).astype(float32)
if not skipflat and not dynamic_ff:
test_im = ring_correction(test_im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor, ext_fov).astype(float32)
else:
test_im = ring_correction(test_im, ringrem, False, False, half_half,
half_half_line / decim_factor, ext_fov).astype(float32)
# Now we can allocate memory for the bunch of slices:
tmp_im = empty((approx_win, test_im.shape[0], test_im.shape[1]), dtype=float32)
tmp_im[0,:,:] = test_im
# Reading all the the sinos from TDF file and close:
for ct in range(1, approx_win):
# Read the sinogram:
test_im = tdf.read_sino(dset, zrange[ct]*downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing for each sinogram of the bunch:
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
#test_im = dynamic_flat_fielding(test_im, zrange[ct] / downsc_factor, EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
test_im = dynamic_flat_fielding(test_im, zrange[ct], EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
else:
#test_im = flat_fielding(test_im, zrange[ct] / downsc_factor, corr_plan, flat_end, half_half,
# half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
test_im = flat_fielding(test_im, zrange[ct], corr_plan, flat_end, half_half,
half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(test_im, ext_fov_rot_right, ext_fov_overlap / downsc_factor, ext_fov_normalize, ext_fov_average).astype(float32)
if not skipflat and not dynamic_ff:
test_im = ring_correction(test_im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor, ext_fov).astype(float32)
else:
test_im = ring_correction(test_im, ringrem, False, False, half_half,
half_half_line / decim_factor, ext_fov).astype(float32)
tmp_im[ct,:,:] = test_im
f_in.close()
# Now everything has to refer to a downscaled dataset:
sino_idx = ((zrange == sino_idx).nonzero())
#
# Perform phase retrieval:
#
# Prepare the plan:
if (phrtmethod == 0):
# Paganin's:
phrtplan = tiehom_plan(tmp_im[:,0,:], phrt_param1, phrt_param2, energy, distance, pixsize * downsc_factor, phrtpad)
else:
phrtplan = phrt_plan(tmp_im[:,0,:], energy, distance, pixsize * downsc_factor, phrt_param2, phrt_param1, phrtmethod, phrtpad)
#phrtplan = prepare_plan (tmp_im[:,0,:], beta, delta, energy, distance,
#pixsize*downsc_factor, padding=phrtpad)
# Process each projection (whose height depends on the size of the bunch):
for ct in range(0, tmp_im.shape[1]):
#tmp_im[:,ct,:] = phase_retrieval(tmp_im[:,ct,:], phrtplan).astype(float32)
if (phrtmethod == 0):
tmp_im[:,ct,:] = tiehom(tmp_im[:,ct,:], phrtplan).astype(float32)
else:
tmp_im[:,ct,:] = phrt(tmp_im[:,ct,:], phrtplan, phrtmethod).astype(float32)
# Extract the requested sinogram:
im = tmp_im[sino_idx[0],:,:].squeeze()
else:
# Read only one sinogram:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
im = tdf.read_sino(dset,sino_idx * downsc_factor).astype(float32)
f_in.close()
# Apply projection removal (if required):
im = im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
im = im[::decim_factor,::downsc_factor]
#sino_idx = sino_idx / downsc_factor # Downscaling for the index already applied
# Perform the preprocessing of the sinogram (if required):
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
im = dynamic_flat_fielding(im, sino_idx, EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
else:
im = flat_fielding(im, sino_idx, corr_plan, flat_end, half_half, half_half_line / decim_factor,
norm_sx, norm_dx).astype(float32)
if ext_fov:
im = extfov_correction(im, ext_fov_rot_right, ext_fov_overlap / downsc_factor, ext_fov_normalize, ext_fov_average)
if not skipflat and not dynamic_ff:
im = ring_correction(im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor, ext_fov)
else:
im = ring_correction(im, ringrem, False, False, half_half,
half_half_line / decim_factor, ext_fov)
# Additional ring removal before reconstruction:
#im = boinhaibel(im, '11;')
#im = munchetal(im, '5;1.8')
#im = rivers(im, '13;')
#im = raven(im, '11;0.8')
#im = oimoen(im, '51;51')
# Actual reconstruction:
im = reconstruct(im, angles, offset / downsc_factor, logtransform, recpar, circle, scale, pad, method,
zerone_mode, dset_min, dset_max, corr_offset, rolling, roll_shift, tmppath).astype(float32)
# Apply post-processing (if required):
if postprocess_required:
im = polarfilter(im, polarfilt_opt)
im = croprescale(im, convert_opt, crop_opt)
else:
# Create the circle mask for fancy output:
if (circle == True):
siz = im.shape[1]
if siz % 2:
rang = arange(-siz / 2 + 1, siz / 2 + 1)
else:
rang = arange(-siz / 2,siz / 2)
x,y = meshgrid(rang,rang)
z = x ** 2 + y ** 2
a = (z < (siz / 2 - int(round(abs(offset) / downsc_factor))) ** 2)
im = im * a
# Write down reconstructed preview file (file name modified with metadata):
im = im.astype(float32)
outfile = outfile + '_' + str(im.shape[1]) + 'x' + str(im.shape[0]) + '_' + str(amin(im)) + '$' + str(amax(im))
im.tofile(outfile)
def process(sino_idx, num_sinos, infile, outpath, preprocessing_required, corr_plan, norm_sx, norm_dx, flat_end, half_half, half_half_line, ext_fov, ext_fov_rot_right, ext_fov_overlap, ext_fov_normalize, ext_fov_average, ringrem, phaseretrieval_required, phrtmethod, phrt_param1, phrt_param2, energy, distance, pixsize, phrtpad, approx_win, angles, offset, logtransform, param1, circle, scale, pad, method, zerone_mode, dset_min, dset_max, decim_factor, downsc_factor, corr_offset, postprocess_required, convert_opt, crop_opt, nr_threads, angles_from, angles_to, logfilename, lock, slice_prefix): """To do... """ slice_nr = sino_idx # Perform reconstruction (on-the-fly preprocessing and phase retrieval, if required): if (phaseretrieval_required): # In this case a bunch of sinograms is loaded into memory: # # Load the temporary data structure reading the input TDF file. # To know the right dimension the first sinogram is pre-processed. # # Open the TDF file and get the dataset: f_in = getHDF5(infile, 'r') if "/tomo" in f_in: dset = f_in['tomo'] else: dset = f_in['exchange/data'] # Downscaling and decimation factors considered when determining the approximation window: zrange = arange(sino_idx - approx_win*downsc_factor/2, sino_idx + approx_win*downsc_factor/2, downsc_factor) zrange = zrange[ (zrange >= 0) ] zrange = zrange[ (zrange < num_sinos) ] approx_win = zrange.shape[0] # Approximation window cannot be odd: if (approx_win % 2 == 1): approx_win = approx_win-1 zrange = zrange[0:approx_win] # Read one sinogram to get the proper dimensions: test_im = tdf.read_sino(dset, zrange[0]*downsc_factor).astype(float32) # Apply decimation and downscaling (if required): test_im = test_im[::decim_factor, ::downsc_factor] # Perform the pre-processing of the first sinogram to get the right dimension: if (preprocessing_required): test_im = flat_fielding (test_im, zrange[0], corr_plan, flat_end, half_half, half_half_line/decim_factor, norm_sx, norm_dx).astype(float32) if ext_fov: test_im = extfov_correction (test_im, ext_fov_rot_right, ext_fov_overlap/downsc_factor, ext_fov_normalize, ext_fov_average).astype(float32) test_im = ring_correction (test_im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half, half_half_line/decim_factor, ext_fov).astype(float32) # Now we can allocate memory for the bunch of slices: tmp_im = empty((approx_win, test_im.shape[0], test_im.shape[1]), dtype=float32) tmp_im[0,:,:] = test_im # Reading all the the sinos from TDF file and close: for ct in range(1, approx_win): test_im = tdf.read_sino(dset, zrange[ct]*downsc_factor).astype(float32) test_im = test_im[::decim_factor, ::downsc_factor] # Perform the pre-processing for each sinogram of the bunch: if (preprocessing_required): test_im = flat_fielding (test_im, zrange[ct], corr_plan, flat_end, half_half, half_half_line/decim_factor, norm_sx, norm_dx).astype(float32) if ext_fov: test_im = extfov_correction (test_im, ext_fov_rot_right, ext_fov_overlap/downsc_factor, ext_fov_normalize, ext_fov_average).astype(float32) test_im = ring_correction (test_im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half, half_half_line/decim_factor, ext_fov).astype(float32) tmp_im[ct,:,:] = test_im f_in.close() # Now everything has to refer to a downscaled dataset: sino_idx = ((zrange == sino_idx).nonzero()) # # Perform phase retrieval: # # Prepare the plan: if (phrtmethod == 0): # Paganin's: phrtplan = tiehom_plan (tmp_im[:,0,:], phrt_param1, phrt_param2, energy, distance, pixsize*downsc_factor, padding=phrtpad) else: phrtplan = phrt_plan (tmp_im[:,0,:], energy, distance, pixsize*downsc_factor, phrt_param2, phrt_param1, phrtmethod, padding=phrtpad) # Process each projection (whose height depends on the size of the bunch): for ct in range(0, tmp_im.shape[1]): if (phrtmethod == 0): tmp_im[:,ct,:] = tiehom(tmp_im[:,ct,:], phrtplan).astype(float32) else: tmp_im[:,ct,:] = phrt(tmp_im[:,ct,:], phrtplan, phrtmethod).astype(float32) # Extract the requested sinogram: im = tmp_im[sino_idx[0],:,:].squeeze() else: # Read only one sinogram: f_in = getHDF5(infile, 'r') if "/tomo" in f_in: dset = f_in['tomo'] else: dset = f_in['exchange/data'] im = tdf.read_sino(dset,sino_idx * downsc_factor).astype(float32) f_in.close() # Downscale and decimate the sinogram: im = im[::decim_factor,::downsc_factor] #sino_idx = sino_idx/downsc_factor # Perform the preprocessing of the sinogram (if required): if (preprocessing_required): im = flat_fielding (im, sino_idx, corr_plan, flat_end, half_half, half_half_line/decim_factor, norm_sx, norm_dx).astype(float32) if ext_fov: test_im = extfov_correction (test_im, ext_fov_rot_right, ext_fov_overlap/downsc_factor, ext_fov_normalize, ext_fov_average).astype(float32) im = ring_correction (im, ringrem, flat_end, corr_plan['skip_flat_after'], half_half, half_half_line/decim_factor, ext_fov) # Log infos: log = open(logfilename,"a") log.write(linesep + "\tPerforming reconstruction with multiple centers of rotation...") log.write(linesep + "\t--------------") log.close() # Split the computation into multiple processes: for num in range(nr_threads): start = ( (angles_to - angles_from + 1) / nr_threads)*num + angles_from if (num == nr_threads - 1): end = angles_to else: end = ( (angles_to - angles_from + 1) / nr_threads)*(num + 1) + angles_from - 1 Process(target=reconstruct, args=(im, angles, offset/downsc_factor, logtransform, param1, circle, scale, pad, method, zerone_mode, dset_min, dset_max, corr_offset, postprocess_required, convert_opt, crop_opt, start, end, outpath, slice_nr, downsc_factor, decim_factor, logfilename, lock, slice_prefix)).start()
def main(argv):
"""To do...
"""
lock = Lock()
skip_flat = True
first_done = False
pyfftw_cache_disable()
pyfftw_cache_enable()
pyfftw_set_keepalive_time(1800)
# Get the from and to number of files to process:
idx = int(argv[0])
# Get full paths of input TDF and output TDF:
infile = argv[1]
outfile = argv[2]
# Get the phase retrieval parameters:
method = int(argv[3])
param1 = double(argv[4]) # param1( e.g. regParam, or beta)
param2 = double(argv[5]) # param2( e.g. thresh or delta)
energy = double(argv[6])
distance = double(argv[7])
pixsize = double(argv[8]) / 1000.0 # pixsixe from micron to mm:
pad = True if argv[9] == "True" else False
# Tmp path and log file:
tmppath = argv[10]
if not tmppath.endswith(sep): tmppath += sep
logfilename = argv[11]
# Open the HDF5 file and check it contains flat files:
skipflat = False
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
if not "/flat" in f_in:
skipflat = True
else:
dset = f_in['exchange/data']
if not "/exchange/data_white" in f_in:
skipflat = True
num_proj = tdf.get_nr_projs(dset)
num_sinos = tdf.get_nr_sinos(dset)
# Check if the HDF5 makes sense:
if (num_proj == 0):
log = open(logfilename, "a")
log.write(linesep + "\tNo projections found. Process will end.")
log.close()
exit()
# Get flats and darks from cache or from file:
if not skipflat:
try:
corrplan = cache2plan(infile, tmppath)
except Exception as e:
#print "Error(s) when reading from cache"
corrplan = extract_flatdark(f_in, True, logfilename)
remove(logfilename)
plan2cache(corrplan, infile, tmppath)
# Read projection:
im = tdf.read_tomo(dset, idx).astype(float32)
f_in.close()
# Apply simple flat fielding (if applicable):
if not skipflat:
if (isinstance(corrplan['im_flat_after'], ndarray)
and isinstance(corrplan['im_flat'], ndarray)
and isinstance(corrplan['im_dark'], ndarray)
and isinstance(corrplan['im_dark_after'], ndarray)):
if (idx < num_proj / 2):
im = (im - corrplan['im_dark']) / (
abs(corrplan['im_flat'] - corrplan['im_dark']) +
finfo(float32).eps)
else:
im = (im - corrplan['im_dark_after']) / (
abs(corrplan['im_flat_after'] - corrplan['im_dark_after'])
+ finfo(float32).eps)
# Prepare plan:
im = im.astype(float32)
if (method == 0):
# Paganin 2002:
plan = tiehom_plan(im, param1, param2, energy, distance, pixsize, pad)
im = tiehom(im, plan).astype(float32)
elif (method == 1):
# Paganin 2020:
plan = tiehom_plan2020(im, param1, param2, energy, distance, pixsize,
pad)
im = tiehom2020(im, plan).astype(float32)
else:
plan = phrt_plan(im, energy, distance, pixsize, param2, param1, method,
pad)
im = phrt(im, plan, method).astype(float32)
# Write down reconstructed preview file (file name modified with metadata):
im = im.astype(float32)
outfile = outfile + '_' + str(im.shape[1]) + 'x' + str(
im.shape[0]) + '_' + str(nanmin(im)) + '$' + str(nanmax(im))
im.tofile(outfile)
def process(sino_idx, num_sinos, infile, outfile, preprocessing_required,
corr_plan, skipflat, norm_sx, norm_dx, flat_end, half_half,
half_half_line, ext_fov, ext_fov_rot_right, ext_fov_overlap,
ext_fov_normalize, ext_fov_average, ringrem,
phaseretrieval_required, phrtmethod, phrt_param1, phrt_param2,
energy, distance, pixsize, phrtpad, approx_win, angles,
angles_projfrom, angles_projto, offset, logtransform, recpar,
circle, scale, pad, method, rolling, roll_shift, zerone_mode,
dset_min, dset_max, decim_factor, downsc_factor, corr_offset,
postprocess_required, polarfilt_opt, convert_opt, crop_opt,
dynamic_ff, EFF, filtEFF, im_dark, nr_threads, logfilename,
tmppath):
"""To do...
"""
# Perform reconstruction (on-the-fly preprocessing and phase retrieval, if
# required):
if (phaseretrieval_required):
# In this case a bunch of sinograms is loaded into memory:
#
# Load the temporary data structure reading the input TDF file.
# To know the right dimension the first sinogram is pre-processed.
#
# Open the TDF file and get the dataset:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
# Downscaling and decimation factors considered when determining the
# approximation window:
zrange = arange(sino_idx - approx_win * downsc_factor / 2,
sino_idx + approx_win * downsc_factor / 2,
downsc_factor)
zrange = zrange[(zrange >= 0)]
zrange = zrange[(zrange < num_sinos)]
approx_win = zrange.shape[0]
# Approximation window cannot be odd:
if (approx_win % 2 == 1):
approx_win = approx_win - 1
zrange = zrange[0:approx_win]
# Read one sinogram to get the proper dimensions:
test_im = tdf.read_sino(dset,
zrange[0] * downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing of the first sinogram to get the right
# dimension:
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
#test_im = dynamic_flat_fielding(test_im, zrange[0] / downsc_factor, EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
test_im = dynamic_flat_fielding(test_im, zrange[0], EFF,
filtEFF, 2, im_dark,
norm_sx, norm_dx)
else:
#test_im = flat_fielding(test_im, zrange[0] / downsc_factor, corr_plan, flat_end, half_half,
# half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
test_im = flat_fielding(test_im, zrange[0], corr_plan,
flat_end, half_half,
half_half_line / decim_factor,
norm_sx, norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(test_im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor,
ext_fov_normalize,
ext_fov_average).astype(float32)
if not skipflat and not dynamic_ff:
test_im = ring_correction(test_im, ringrem, flat_end,
corr_plan['skip_flat_after'],
half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
else:
test_im = ring_correction(test_im, ringrem, False, False,
half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
# Now we can allocate memory for the bunch of slices:
tmp_im = empty((approx_win, test_im.shape[0], test_im.shape[1]),
dtype=float32)
tmp_im[0, :, :] = test_im
# Reading all the the sinos from TDF file and close:
for ct in range(1, approx_win):
# Read the sinogram:
test_im = tdf.read_sino(dset,
zrange[ct] * downsc_factor).astype(float32)
# Apply projection removal (if required):
test_im = test_im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
test_im = test_im[::decim_factor, ::downsc_factor]
# Perform the pre-processing for each sinogram of the bunch:
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
#test_im = dynamic_flat_fielding(test_im, zrange[ct] / downsc_factor, EFF, filtEFF, 2, im_dark, norm_sx, norm_dx)
test_im = dynamic_flat_fielding(
test_im, zrange[ct], EFF, filtEFF, 2, im_dark,
norm_sx, norm_dx)
else:
#test_im = flat_fielding(test_im, zrange[ct] / downsc_factor, corr_plan, flat_end, half_half,
# half_half_line / decim_factor, norm_sx, norm_dx).astype(float32)
test_im = flat_fielding(test_im, zrange[ct], corr_plan,
flat_end, half_half,
half_half_line / decim_factor,
norm_sx,
norm_dx).astype(float32)
if ext_fov:
test_im = extfov_correction(
test_im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor, ext_fov_normalize,
ext_fov_average).astype(float32)
if not skipflat and not dynamic_ff:
test_im = ring_correction(test_im, ringrem, flat_end,
corr_plan['skip_flat_after'],
half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
else:
test_im = ring_correction(test_im, ringrem, False, False,
half_half,
half_half_line / decim_factor,
ext_fov).astype(float32)
tmp_im[ct, :, :] = test_im
f_in.close()
# Now everything has to refer to a downscaled dataset:
sino_idx = ((zrange == sino_idx).nonzero())
#
# Perform phase retrieval:
#
# Prepare the plan:
if (phrtmethod == 0):
# Paganin 2002:
phrtplan = tiehom_plan(tmp_im[:, 0, :], phrt_param1, phrt_param2,
energy, distance, pixsize * downsc_factor,
phrtpad)
elif (phrtmethod == 1):
# Paganin 2020:
phrtplan = tiehom_plan2020(tmp_im[:, 0, :], phrt_param1,
phrt_param2, energy, distance,
pixsize * downsc_factor, phrtpad)
else:
phrtplan = phrt_plan(tmp_im[:, 0, :], energy, distance,
pixsize * downsc_factor, phrt_param2,
phrt_param1, phrtmethod, phrtpad)
#phrtplan = prepare_plan (tmp_im[:,0,:], beta, delta, energy, distance,
#pixsize*downsc_factor, padding=phrtpad)
# Process each projection (whose height depends on the size of the bunch):
for ct in range(0, tmp_im.shape[1]):
#tmp_im[:,ct,:] = phase_retrieval(tmp_im[:,ct,:], phrtplan).astype(float32)
if (phrtmethod == 0):
tmp_im[:, ct, :] = tiehom(tmp_im[:, ct, :],
phrtplan).astype(float32)
elif (phrtmethod == 1):
tmp_im[:, ct, :] = tiehom2020(tmp_im[:, ct, :],
phrtplan).astype(float32)
else:
tmp_im[:, ct, :] = phrt(tmp_im[:, ct, :], phrtplan,
phrtmethod).astype(float32)
# Extract the requested sinogram:
im = tmp_im[sino_idx[0], :, :].squeeze()
else:
# Read only one sinogram:
f_in = getHDF5(infile, 'r')
if "/tomo" in f_in:
dset = f_in['tomo']
else:
dset = f_in['exchange/data']
im = tdf.read_sino(dset, sino_idx * downsc_factor).astype(float32)
f_in.close()
# Apply projection removal (if required):
im = im[angles_projfrom:angles_projto, :]
# Apply decimation and downscaling (if required):
im = im[::decim_factor, ::downsc_factor]
#sino_idx = sino_idx / downsc_factor # Downscaling for the index already applied
# Perform the preprocessing of the sinogram (if required):
if (preprocessing_required):
if not skipflat:
if dynamic_ff:
# Dynamic flat fielding with downsampling = 2:
im = dynamic_flat_fielding(im, sino_idx, EFF, filtEFF, 2,
im_dark, norm_sx, norm_dx)
else:
im = flat_fielding(im, sino_idx, corr_plan, flat_end,
half_half,
half_half_line / decim_factor, norm_sx,
norm_dx).astype(float32)
if ext_fov:
im = extfov_correction(im, ext_fov_rot_right,
ext_fov_overlap / downsc_factor,
ext_fov_normalize, ext_fov_average)
if not skipflat and not dynamic_ff:
im = ring_correction(im, ringrem, flat_end,
corr_plan['skip_flat_after'], half_half,
half_half_line / decim_factor, ext_fov)
else:
im = ring_correction(im, ringrem, False, False, half_half,
half_half_line / decim_factor, ext_fov)
# Additional ring removal before reconstruction:
#im = boinhaibel(im, '11;')
#im = munchetal(im, '5;1.8')
#im = rivers(im, '13;')
#im = raven(im, '11;0.8')
#im = oimoen(im, '51;51')
# Actual reconstruction:
im = reconstruct(im, angles, offset / downsc_factor, logtransform, recpar,
circle, scale, pad, method, zerone_mode, dset_min,
dset_max, corr_offset, rolling, roll_shift,
tmppath).astype(float32)
# Apply post-processing (if required):
if postprocess_required:
im = polarfilter(im, polarfilt_opt)
im = croprescale(im, convert_opt, crop_opt)
else:
# Create the circle mask for fancy output:
if (circle == True):
siz = im.shape[1]
if siz % 2:
rang = arange(-siz / 2 + 1, siz / 2 + 1)
else:
rang = arange(-siz / 2, siz / 2)
x, y = meshgrid(rang, rang)
z = x**2 + y**2
a = (z < (siz / 2 - int(round(abs(offset) / downsc_factor)))**2)
im = im * a
# Write down reconstructed preview file (file name modified with metadata):
im = im.astype(float32)
outfile = outfile + '_' + str(im.shape[1]) + 'x' + str(
im.shape[0]) + '_' + str(amin(im)) + '$' + str(amax(im))
im.tofile(outfile)
def main(argv): """To do... """ lock = Lock() skip_flat = True first_done = False pyfftw_cache_disable() pyfftw_cache_enable() pyfftw_set_keepalive_time(1800) # Get the from and to number of files to process: idx = int(argv[0]) # Get full paths of input TDF and output TDF: infile = argv[1] outfile = argv[2] # Get the phase retrieval parameters: method = int(argv[3]) param1 = double(argv[4]) # param1( e.g. regParam, or beta) param2 = double(argv[5]) # param2( e.g. thresh or delta) energy = double(argv[6]) distance = double(argv[7]) pixsize = double(argv[8]) / 1000.0 # pixsixe from micron to mm: pad = True if argv[9] == "True" else False # Tmp path and log file: tmppath = argv[10] if not tmppath.endswith(sep): tmppath += sep logfilename = argv[11] # Open the HDF5 file and check it contains flat files: skipflat = False f_in = getHDF5(infile, 'r') if "/tomo" in f_in: dset = f_in['tomo'] if not "/flat" in f_in: skipflat = True else: dset = f_in['exchange/data'] if not "/exchange/data_white" in f_in: skipflat = True num_proj = tdf.get_nr_projs(dset) num_sinos = tdf.get_nr_sinos(dset) # Check if the HDF5 makes sense: if (num_proj == 0): log = open(logfilename,"a") log.write(linesep + "\tNo projections found. Process will end.") log.close() exit() # Get flats and darks from cache or from file: if not skipflat: try: corrplan = cache2plan(infile, tmppath) except Exception as e: #print "Error(s) when reading from cache" corrplan = extract_flatdark(f_in, True, logfilename) remove(logfilename) plan2cache(corrplan, infile, tmppath) # Read projection: im = tdf.read_tomo(dset,idx).astype(float32) f_in.close() # Apply simple flat fielding (if applicable): if not skipflat: if (isinstance(corrplan['im_flat_after'], ndarray) and isinstance(corrplan['im_flat'], ndarray) and isinstance(corrplan['im_dark'], ndarray) and isinstance(corrplan['im_dark_after'], ndarray)) : if (idx < num_proj/2): im = (im - corrplan['im_dark']) / (abs(corrplan['im_flat'] - corrplan['im_dark']) + finfo(float32).eps) else: im = (im - corrplan['im_dark_after']) / (abs(corrplan['im_flat_after'] - corrplan['im_dark_after']) + finfo(float32).eps) # Prepare plan: im = im.astype(float32) if (method == 0): # Paganin's: plan = tiehom_plan (im, param1, param2, energy, distance, pixsize, pad) im = tiehom(im, plan).astype(float32) else: plan = phrt_plan (im, energy, distance, pixsize, param2, param1, method, pad) im = phrt(im, plan, method).astype(float32) # Write down reconstructed preview file (file name modified with metadata): im = im.astype(float32) outfile = outfile + '_' + str(im.shape[1]) + 'x' + str(im.shape[0]) + '_' + str( nanmin(im)) + '$' + str( nanmax(im) ) im.tofile(outfile)