def main():
parser = _build_arg_parser()
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
if not nib.streamlines.is_supported(args.out_tractogram):
parser.error('Invalid output streamline file format (must be trk or ' +
'tck): {0}'.format(args.out_tractogram))
inputs = [args.in_odf, args.in_seed, args.in_mask]
assert_inputs_exist(parser, inputs)
assert_outputs_exist(parser, args, args.out_tractogram)
verify_streamline_length_options(parser, args)
verify_compression_th(args.compress)
verify_seed_options(parser, args)
theta = gm.math.radians(get_theta(args.theta, args.algo))
max_nbr_pts = int(args.max_length / args.step_size)
min_nbr_pts = int(args.min_length / args.step_size) + 1
max_invalid_dirs = int(math.ceil(args.max_invalid_length / args.step_size))
logging.debug("Loading seeding mask.")
seed_img = nib.load(args.in_seed)
seed_data = seed_img.get_fdata(caching='unchanged', dtype=float)
seed_res = seed_img.header.get_zooms()[:3]
seed_generator = SeedGenerator(seed_data, seed_res)
if args.npv:
# toDo. This will not really produce n seeds per voxel, only true
# in average.
nbr_seeds = len(seed_generator.seeds) * args.npv
elif args.nt:
nbr_seeds = args.nt
else:
# Setting npv = 1.
nbr_seeds = len(seed_generator.seeds)
if len(seed_generator.seeds) == 0:
parser.error(
'Seed mask "{}" does not have any voxel with value > 0.'.format(
args.in_seed))
logging.debug("Loading tracking mask.")
mask_img = nib.load(args.in_mask)
mask_data = mask_img.get_fdata(caching='unchanged', dtype=float)
mask_res = mask_img.header.get_zooms()[:3]
mask = DataVolume(mask_data, mask_res, args.mask_interp)
logging.debug("Loading ODF SH data.")
odf_sh_img = nib.load(args.in_odf)
odf_sh_data = odf_sh_img.get_fdata(caching='unchanged', dtype=float)
odf_sh_res = odf_sh_img.header.get_zooms()[:3]
dataset = DataVolume(odf_sh_data, odf_sh_res, args.sh_interp)
logging.debug("Instantiating propagator.")
propagator = ODFPropagator(dataset, args.step_size, args.rk_order,
args.algo, args.sh_basis, args.sf_threshold,
args.sf_threshold_init, theta, args.sphere)
logging.debug("Instantiating tracker.")
tracker = Tracker(propagator,
mask,
seed_generator,
nbr_seeds,
min_nbr_pts,
max_nbr_pts,
max_invalid_dirs,
compression_th=args.compress,
nbr_processes=args.nbr_processes,
save_seeds=args.save_seeds,
mmap_mode='r+',
rng_seed=args.rng_seed,
track_forward_only=args.forward_only,
skip=args.skip)
start = time.time()
logging.debug("Tracking...")
streamlines, seeds = tracker.track()
str_time = "%.2f" % (time.time() - start)
logging.debug("Tracked {} streamlines (out of {} seeds), in {} seconds.\n"
"Now saving...".format(len(streamlines), nbr_seeds,
str_time))
# save seeds if args.save_seeds is given
data_per_streamline = {'seeds': seeds} if args.save_seeds else {}
# Silencing SFT's logger if our logging is in DEBUG mode, because it
# typically produces a lot of outputs!
set_sft_logger_level('WARNING')
# Compared with scil_compute_local_tracking, using sft rather than
# LazyTractogram to deal with space.
# Contrary to scilpy or dipy, where space after tracking is vox, here
# space after tracking is voxmm.
# Smallest possible streamline coordinate is (0,0,0), equivalent of
# corner origin (TrackVis)
sft = StatefulTractogram(streamlines,
mask_img,
Space.VOXMM,
Origin.TRACKVIS,
data_per_streamline=data_per_streamline)
save_tractogram(sft, args.out_tractogram)
def main():
parser = _build_arg_parser()
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
assert_inputs_exist(parser, [args.in_odf, args.in_seed, args.in_mask])
assert_outputs_exist(parser, args, args.out_tractogram)
if not nib.streamlines.is_supported(args.out_tractogram):
parser.error('Invalid output streamline file format (must be trk or ' +
'tck): {0}'.format(args.out_tractogram))
verify_streamline_length_options(parser, args)
verify_compression_th(args.compress)
verify_seed_options(parser, args)
mask_img = nib.load(args.in_mask)
mask_data = get_data_as_mask(mask_img, dtype=bool)
# Make sure the data is isotropic. Else, the strategy used
# when providing information to dipy (i.e. working as if in voxel space)
# will not yield correct results.
odf_sh_img = nib.load(args.in_odf)
if not np.allclose(np.mean(odf_sh_img.header.get_zooms()[:3]),
odf_sh_img.header.get_zooms()[0], atol=1e-03):
parser.error(
'ODF SH file is not isotropic. Tracking cannot be ran robustly.')
if args.npv:
nb_seeds = args.npv
seed_per_vox = True
elif args.nt:
nb_seeds = args.nt
seed_per_vox = False
else:
nb_seeds = 1
seed_per_vox = True
voxel_size = odf_sh_img.header.get_zooms()[0]
vox_step_size = args.step_size / voxel_size
seed_img = nib.load(args.in_seed)
seeds = track_utils.random_seeds_from_mask(
seed_img.get_fdata(dtype=np.float32),
np.eye(4),
seeds_count=nb_seeds,
seed_count_per_voxel=seed_per_vox,
random_seed=args.seed)
# Tracking is performed in voxel space
max_steps = int(args.max_length / args.step_size) + 1
streamlines_generator = LocalTracking(
_get_direction_getter(args),
BinaryStoppingCriterion(mask_data),
seeds, np.eye(4),
step_size=vox_step_size, max_cross=1,
maxlen=max_steps,
fixedstep=True, return_all=True,
random_seed=args.seed,
save_seeds=args.save_seeds)
scaled_min_length = args.min_length / voxel_size
scaled_max_length = args.max_length / voxel_size
if args.save_seeds:
filtered_streamlines, seeds = \
zip(*((s, p) for s, p in streamlines_generator
if scaled_min_length <= length(s) <= scaled_max_length))
data_per_streamlines = {'seeds': lambda: seeds}
else:
filtered_streamlines = \
(s for s in streamlines_generator
if scaled_min_length <= length(s) <= scaled_max_length)
data_per_streamlines = {}
if args.compress:
filtered_streamlines = (
compress_streamlines(s, args.compress)
for s in filtered_streamlines)
tractogram = LazyTractogram(lambda: filtered_streamlines,
data_per_streamlines,
affine_to_rasmm=seed_img.affine)
filetype = nib.streamlines.detect_format(args.out_tractogram)
reference = get_reference_info(seed_img)
header = create_tractogram_header(filetype, *reference)
# Use generator to save the streamlines on-the-fly
nib.streamlines.save(tractogram, args.out_tractogram, header=header)
def main():
parser = build_argparser()
args = parser.parse_args()
logging.basicConfig(level=args.logging.upper())
# ----- Checks
if not nib.streamlines.is_supported(args.out_tractogram):
parser.error('Invalid output streamline file format (must be trk or '
'tck): {0}'.format(args.out_tractogram))
assert_inputs_exist(parser, args.hdf5_file)
assert_outputs_exist(parser, args, args.out_tractogram)
verify_streamline_length_options(parser, args)
verify_compression_th(args.compress)
verify_seed_options(parser, args)
# ----- Prepare values
max_nbr_pts = int(args.max_length / args.step_size)
min_nbr_pts = int(args.min_length / args.step_size) + 1
max_invalid_dirs = int(math.ceil(args.max_invalid_len / args.step_size))
# r+ is necessary for interpolation function in cython who need read/write
# rights
mmap_mode = None if args.set_mmap_to_none else 'r+'
device = torch.device('cpu')
if args.use_gpu:
if args.nbr_processes > 1:
logging.warning("Number of processes was set to {} but you "
"are using GPU. Parameter ignored.".format(
args.nbr_processes))
if torch.cuda.is_available():
device = torch.device('cuda')
hdf_handle = h5py.File(args.hdf5_file, 'r')
tracker, ref = prepare_tracker(parser, args, hdf_handle, device,
min_nbr_pts, max_nbr_pts, max_invalid_dirs,
mmap_mode)
# ----- Track
with Timer("\nTracking...", newline=True, color='blue'):
streamlines, seeds = tracker.track()
logging.debug(
"Tracked {} streamlines (out of {} seeds). Now saving...".format(
len(streamlines), tracker.nbr_seeds))
# save seeds if args.save_seeds is given
data_per_streamline = {'seed': lambda: seeds} if args.save_seeds else {}
# Silencing SFT's logger if our logging is in DEBUG mode, because it
# typically produces a lot of outputs!
set_sft_logger_level('WARNING')
sft = StatefulTractogram(streamlines,
ref,
Space.VOXMM,
data_per_streamline=data_per_streamline)
save_tractogram(sft, args.out_tractogram, bbox_valid_check=False)