def update_paths(args, upper_dirs=None, pattern='path'):
""" find params with not existing paths
:param dict args: dictionary with all parameters
:param list(str) upper_dirs: list of keys in parameters
with item for which only the parent folder must exist
:param str pattern: patter specifying key with path
:return list(str): key of missing paths
>>> update_paths({'sample': 123})[1]
[]
>>> update_paths({'path_': '.'})[1]
[]
>>> params = {'path_out': './nothing'}
>>> update_paths(params)[1]
['path_out']
>>> update_paths(params, upper_dirs=['path_out'])[1]
[]
"""
if upper_dirs is None:
upper_dirs = []
missing = []
for k in (k for k in args if pattern in k):
if '*' in os.path.basename(args[k]) or k in upper_dirs:
p = update_path(os.path.dirname(args[k]))
args[k] = os.path.join(p, os.path.basename(args[k]))
else:
args[k] = update_path(args[k])
p = args[k]
if not os.path.exists(p):
logging.warning('missing "%s": %s', k, p)
missing.append(k)
return args, missing
def filter_paired_landmarks(item, path_dataset, path_reference, col_source,
col_target):
""" filter all relevant landmarks which were used and copy them to experiment
The case is that in certain challenge stage users had provided just a subset
of all image landmarks which could be laos shuffled. The idea is to filter identify
all user used (provided in dataset) landmarks and filter them from temporary
reference dataset.
:param dict|Series item: experiment DataFrame
:param str path_dataset: path to provided landmarks
:param str path_reference: path to the complete landmark collection
:param str col_source: column name of landmarks to be transformed
:param str col_target: column name of landmarks to be compared
:return tuple(float,ndarray,ndarray): match ratio, filtered ref and move landmarks
>>> p_data = update_path('data-images')
>>> p_csv = os.path.join(p_data, 'pairs-imgs-lnds_histol.csv')
>>> df = pd.read_csv(p_csv)
>>> ratio, lnds_ref, lnds_move = filter_paired_landmarks(dict(df.iloc[0]), p_data, p_data,
... ImRegBenchmark.COL_POINTS_MOVE, ImRegBenchmark.COL_POINTS_REF)
>>> ratio
1.0
>>> lnds_ref.shape == lnds_move.shape
True
"""
path_ref = update_path(item[col_source], pre_path=path_reference)
if not os.path.isfile(path_ref):
raise FileNotFoundError('missing landmarks: %s' % path_ref)
path_load = update_path(item[col_source], pre_path=path_dataset)
if not os.path.isfile(path_load):
raise FileNotFoundError('missing landmarks: %s' % path_load)
pairs = common_landmarks(load_landmarks(path_ref),
load_landmarks(path_load),
threshold=1)
if not pairs.size:
logging.warning(
'there is not pairing between landmarks or dataset and user reference'
)
return 0., np.empty([0]), np.empty([0])
pairs = sorted(pairs.tolist(), key=lambda p: p[1])
ind_ref = np.asarray(pairs)[:, 0]
nb_common = min([
len(load_landmarks(update_path(item[col], pre_path=path_reference)))
for col in (col_target, col_source)
])
ind_ref = ind_ref[ind_ref < nb_common]
path_lnd_ref = update_path(item[col_target], pre_path=path_reference)
lnds_filter_ref = load_landmarks(path_lnd_ref)[ind_ref]
path_lnd_move = update_path(item[col_source], pre_path=path_reference)
lnds_filter_move = load_landmarks(path_lnd_move)[ind_ref]
ratio_matches = len(ind_ref) / float(nb_common)
if ratio_matches > 1:
raise ValueError(
'suspicious ratio for %i paired and %i common landmarks' %
(len(pairs), nb_common))
return ratio_matches, lnds_filter_ref, lnds_filter_move
def replicate_missing_warped_landmarks(df_experiments, path_dataset,
path_experiment):
""" if some warped landmarks are missing replace the path by initial landmarks
:param DF df_experiments: experiment table
:param str path_dataset: path to dataset folder
:param str path_experiment: path ti user experiment folder
:return DF: experiment table
"""
# find empty warped landmarks paths
missing_mask = df_experiments[ImRegBenchmark.COL_POINTS_MOVE_WARP].isnull()
# for the empty place the initial landmarks
df_experiments.loc[missing_mask, ImRegBenchmark.COL_POINTS_MOVE_WARP] = \
df_experiments.loc[missing_mask, ImRegBenchmark.COL_POINTS_MOVE]
# for the empty place maximal execution time
df_experiments.loc[missing_mask, ImRegBenchmark.COL_TIME] = \
df_experiments[ImRegBenchmark.COL_TIME].max()
count = 0
# iterate over whole table
for idx, row in df_experiments.iterrows():
path_csv = update_path(row[ImRegBenchmark.COL_POINTS_MOVE_WARP],
pre_path=path_experiment)
if not os.path.isfile(path_csv):
path_csv = update_path(row[ImRegBenchmark.COL_POINTS_MOVE],
pre_path=path_dataset)
df_experiments.loc[idx,
ImRegBenchmark.COL_POINTS_MOVE_WARP] = path_csv
count += 1
logging.info('Missing warped landmarks: %i', count)
return df_experiments
def _visual_image_move_warp_lnds_move_warp(cls, item, path_dataset=None, path_experiment=None):
""" visualise the case with warped moving image and landmarks
to the reference frame so they are simple to overlap
:param dict item: row with the experiment
:param str|None path_dataset: path to the dataset folder
:param str|None path_experiment: path to the experiment folder
:return obj|None:
"""
assert isinstance(item.get(cls.COL_POINTS_MOVE_WARP, None), str), \
'Missing registered points in "%s"' % cls.COL_POINTS_MOVE_WARP
path_points_warp = update_path(item[cls.COL_POINTS_MOVE_WARP], pre_path=path_experiment)
if not os.path.isfile(path_points_warp):
logging.warning('missing warped landmarks for: %r', dict(item))
return
points_ref, points_move, path_img_ref = cls._load_landmarks(item, path_dataset)
image_warp = cls._load_warped_image(item, path_experiment)
points_warp = load_landmarks(path_points_warp)
if not list(points_warp):
return
# draw image with landmarks
image = draw_image_points(image_warp, points_warp)
_path = update_path(item[cls.COL_REG_DIR], pre_path=path_experiment)
save_image(os.path.join(_path, cls.NAME_IMAGE_MOVE_WARP_POINTS), image)
del image
# visualise the landmarks move during registration
image_ref = load_image(path_img_ref)
fig = draw_images_warped_landmarks(image_ref, image_warp, points_move, points_ref, points_warp)
del image_ref, image_warp
return fig
def filter_export_landmarks(idx_row, path_output, path_dataset,
path_reference):
""" filter all relevant landmarks which were used and copy them to experiment
The case is that in certain challenge stage users had provided just a subset
of all image landmarks which could be laos shuffled. The idea is to filter identify
all user used (provided in dataset) landmarks and filter them from temporary
reference dataset.
:param tuple(idx,dict|Series) idx_row: experiment DataFrame
:param str path_output: path to output folder
:param str path_dataset: path to provided landmarks
:param str path_reference: path to the complete landmark collection
:return tuple(idx,float): record index and match ratio
"""
idx, row = idx_row
ratio_matches, lnds_filter_ref, lnds_filter_move = \
filter_paired_landmarks(row, path_dataset, path_reference,
ImRegBenchmark.COL_POINTS_MOVE,
ImRegBenchmark.COL_POINTS_REF)
# moving and reference landmarks
for col, lnds_flt in [(ImRegBenchmark.COL_POINTS_REF, lnds_filter_ref),
(ImRegBenchmark.COL_POINTS_MOVE, lnds_filter_move)]:
path_out = update_path(row[col], pre_path=path_output)
create_folder(os.path.dirname(path_out), ok_existing=True)
if os.path.isfile(path_out):
assert np.array_equal(load_landmarks(path_out), lnds_flt), \
'overwrite different set of landmarks'
save_landmarks(path_out, lnds_flt)
return idx, ratio_matches
def visualise_registration(cls,
idx_row,
path_dataset=None,
path_experiment=None):
""" visualise the registration results according what landmarks were
estimated - in registration or moving frame
:param tuple(int,dict) df_row: row from iterated table
:param str path_dataset: path to the dataset folder
:param str path_experiment: path to the experiment folder
"""
_, row = idx_row
row = dict(row) # convert even series to dictionary
fig, path_fig = None, None
# visualise particular experiment by idx
if isinstance(row.get(cls.COL_POINTS_MOVE_WARP), str):
fig = cls._visual_image_move_warp_lnds_move_warp(
row, path_dataset, path_experiment)
elif isinstance(row.get(cls.COL_POINTS_REF_WARP), str):
fig = cls._visual_image_move_warp_lnds_ref_warp(
row, path_dataset, path_experiment)
else:
logging.error('Visualisation: no output image or landmarks')
if fig is not None:
path_fig = os.path.join(
update_path(row[cls.COL_REG_DIR], pre_path=path_experiment),
cls.NAME_IMAGE_WARPED_VISUAL)
export_figure(path_fig, fig)
return path_fig
def generate_reg_pairs(rp_imgs,
rp_lnds,
pairs,
public,
path_images=DATASET_IMAGES):
""" format a registration pair as dictionaries/rows in cover table for a set
:param list(str) rp_imgs: relative paths to images
:param rp_lnds: relative paths to related landmarks
:param list(tuple(int,int)) pairs: pairing among images/landmarks
:param list(bool) public: marks whether the particular pair is training or evaluation
:param str path_images: path to the dataset folder
:return list(dict): registration pairs
"""
reg_pairs = []
for k, (i, j) in enumerate(pairs):
img_size, img_diag = image_sizes(
update_path(rp_imgs[i], pre_path=path_images))
reg_pairs.append({
ImRegBenchmark.COL_IMAGE_REF:
rp_imgs[i],
ImRegBenchmark.COL_IMAGE_MOVE:
rp_imgs[j],
ImRegBenchmark.COL_POINTS_REF:
rp_lnds[i],
ImRegBenchmark.COL_POINTS_MOVE:
rp_lnds[j],
ImRegBenchmark.COL_STATUS:
VAL_STATUS_TRAIN if public[k] else VAL_STATUS_TEST,
ImRegBenchmark.COL_IMAGE_SIZE:
img_size,
ImRegBenchmark.COL_IMAGE_DIAGONAL:
img_diag,
})
return reg_pairs
def _visual_image_move_warp_lnds_ref_warp(cls,
item,
path_dataset=None,
path_experiment=None):
""" visualise the case with warped reference landmarks to the move frame
:param dict item: row with the experiment
:param str|None path_dataset: path to the dataset folder
:param str|None path_experiment: path to the experiment folder
:return obj|None:
"""
if not isinstance(item.get(cls.COL_POINTS_REF_WARP), str):
raise ValueError('Missing registered points in "%s"' %
cls.COL_POINTS_REF_WARP)
path_points_warp = update_path(item[cls.COL_POINTS_REF_WARP],
pre_path=path_experiment)
if not os.path.isfile(path_points_warp):
logging.warning('missing warped landmarks for: %r', dict(item))
return
points_ref, points_move, path_img_ref = cls._load_landmarks(
item, path_dataset)
points_warp = load_landmarks(path_points_warp)
if not list(points_warp):
return
# draw image with landmarks
image_move = load_image(
update_path(item[cls.COL_IMAGE_MOVE], pre_path=path_dataset))
image = draw_image_points(image_move, points_warp)
_path = update_path(item[cls.COL_REG_DIR], pre_path=path_experiment)
save_image(os.path.join(_path, cls.NAME_IMAGE_REF_POINTS_WARP), image)
del image
image_ref = load_image(path_img_ref)
image_warp = cls._load_warped_image(item, path_experiment)
image = overlap_two_images(image_ref, image_warp)
_path = update_path(item[cls.COL_REG_DIR], pre_path=path_experiment)
save_image(os.path.join(_path, cls.NAME_IMAGE_REF_WARP), image)
del image, image_warp
# visualise the landmarks move during registration
fig = draw_images_warped_landmarks(image_ref, image_move, points_ref,
points_move, points_warp)
return fig
def update_path_(path, path_base=None):
""" update the image path with possible base path
:param str path: the last path of the path
:param str|None path_base: optional base path
:return str: update path
"""
path = os.path.join(path_base, str(path)) if path_base else path
return update_path(path, absolute=True)
def replicate_missing_warped_landmarks(df_experiments, path_dataset,
path_experiment):
""" if some warped landmarks are missing replace the path by initial landmarks
:param DF df_experiments: experiment table
:param str path_dataset: path to dataset folder
:param str path_experiment: path ti user experiment folder
:return DF: experiment table
"""
# find empty warped landmarks paths
missing_mask = df_experiments[ImRegBenchmark.COL_POINTS_MOVE_WARP].isnull()
if ImRegBenchmark.COL_POINTS_REF_WARP in df_experiments.columns:
# if there ar elaso target warped landmarks, allow to use them
missing_mask &= df_experiments[
ImRegBenchmark.COL_POINTS_REF_WARP].isnull()
# for the empty place the initial landmarks
df_experiments.loc[missing_mask, ImRegBenchmark.COL_POINTS_MOVE_WARP] = \
df_experiments.loc[missing_mask, ImRegBenchmark.COL_POINTS_MOVE]
# for the empty place maximal execution time
df_experiments.loc[missing_mask, ImRegBenchmark.COL_TIME] = \
df_experiments[ImRegBenchmark.COL_TIME].max()
count = 0
# iterate over whole table and check if the path is valid
for idx, row in df_experiments.iterrows():
# select refence/moving warped landmarks
use_move_warp = isinstance(
row.get(ImRegBenchmark.COL_POINTS_MOVE_WARP, None), str)
col_lnds_warp = ImRegBenchmark.COL_POINTS_MOVE_WARP \
if use_move_warp else ImRegBenchmark.COL_POINTS_REF_WARP
# extract the CSV path
path_csv = update_path(row[col_lnds_warp], pre_path=path_experiment)
if not os.path.isfile(path_csv):
# if the path is false, put there the initial from dataset
path_csv = update_path(row[ImRegBenchmark.COL_POINTS_MOVE],
pre_path=path_dataset)
df_experiments.loc[idx,
ImRegBenchmark.COL_POINTS_MOVE_WARP] = path_csv
count += 1
logging.info('Missing warped landmarks: %i', count)
return df_experiments
def _update_path(self, path, destination='data'):
""" update te path to the dataset or output
:param str path: original path
:param str destination: type of update - data | output | general
:return str: updated path
"""
if destination == 'data' and 'path_dataset' in self.params:
path = os.path.join(self.params['path_dataset'], path)
elif destination == 'expt' and 'path_exp' in self.params:
path = os.path.join(self.params['path_exp'], path)
path = update_path(path, absolute=True)
return path
def _absolute_path(self, path, destination='data', base_path=None):
""" update te path to the dataset or output
:param str path: original path
:param str destination: type of update
`data` for data source and `expt` for output experimental folder
:param str destination: type of update
:return str: updated path
"""
if destination and destination == 'data' and 'path_dataset' in self.params:
path = os.path.join(self.params['path_dataset'], path)
elif destination and destination == 'expt' and 'path_exp' in self.params:
path = os.path.join(self.params['path_exp'], path)
path = update_path(path, absolute=True)
return path
def assert_paths(args):
""" check missing paths
:param {} args: dictionary of arguments
:return {}: dictionary of updated arguments
>>> assert_paths({'path_': 'missing'}) # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS
Traceback (most recent call last):
...
AssertionError: missing: (path_) "..."
>>> assert_paths({'abc': 123})
{'abc': 123}
"""
for k in (k for k in args if 'path' in k):
args[k] = update_path(args[k])
assert os.path.exists(args[k]), 'missing: (%s) "%s"' % (k, args[k])
return args
def __check_exist_path(self):
"""Check existence of all paths in parameters.
check existence of all parameters dictionary which has contains words:
'path', 'dir', 'file'
"""
if 'path_out' not in self.params:
raise ValueError('missing "path_out" among parameters')
self.params['path_out'] = update_path(self.params.get('path_out'))
list_names = [n for n in self.params if any(m in n.lower() for m in ['path', 'dir', 'file'])]
for n in list_names:
p = os.path.abspath(os.path.expanduser(self.params[n]))
if not os.path.exists(p):
raise FileNotFoundError('given path/file/dir "%s" does not exist!' % p)
self.params[n] = p
for n in [n for n in self.params if 'exec' in n]:
# in case you define executable in your home
if os.path.expanduser(self.params[n]) != self.params[n]:
self.params[n] = os.path.expanduser(self.params[n])
def _load_warped_image(cls, item, path_experiment=None):
"""load the wapted image if it exists
:param dict item: row with the experiment
:param str|None path_experiment: path to the experiment folder
:return ndarray:
"""
name_img = item.get(cls.COL_IMAGE_MOVE_WARP, None)
if not isinstance(name_img, str):
logging.warning('Missing registered image in "%s"', cls.COL_IMAGE_MOVE_WARP)
image_warp = None
else:
path_img_warp = update_path(name_img, pre_path=path_experiment)
if os.path.isfile(path_img_warp):
image_warp = load_image(path_img_warp)
else:
logging.warning('Define image is missing: %s', path_img_warp)
image_warp = None
return image_warp
def filter_landmarks(idx_row, path_output, path_dataset, path_reference):
""" filter all relevant landmarks which were used and copy them to experiment
:param tuple(idx,dict|Series) idx_row: experiment DataFrame
:param str path_output: path to output folder
:param str path_dataset: path to provided landmarks
:param str path_reference: path to the complete landmark collection
:return tuple(idx,float): record index and match ratio
"""
idx, row = idx_row
path_ref = update_path(row[ImRegBenchmark.COL_POINTS_MOVE],
pre_path=path_reference)
path_load = update_path(row[ImRegBenchmark.COL_POINTS_MOVE],
pre_path=path_dataset)
pairs = common_landmarks(load_landmarks(path_ref),
load_landmarks(path_load),
threshold=1)
if not pairs.size:
return idx, 0.
pairs = sorted(pairs.tolist(), key=lambda p: p[1])
ind_ref = np.asarray(pairs)[:, 0]
nb_common = min([
len(load_landmarks(update_path(row[col], pre_path=path_reference))) for
col in [ImRegBenchmark.COL_POINTS_REF, ImRegBenchmark.COL_POINTS_MOVE]
])
ind_ref = ind_ref[ind_ref < nb_common]
# moving and reference landmarks
for col in [ImRegBenchmark.COL_POINTS_REF, ImRegBenchmark.COL_POINTS_MOVE]:
path_in = update_path(row[col], pre_path=path_reference)
path_out = update_path(row[col], pre_path=path_output)
create_folder(os.path.dirname(path_out), ok_existing=True)
save_landmarks(path_out, load_landmarks(path_in)[ind_ref])
# save ratio of found landmarks
len_lnds_ref = len(
load_landmarks(
update_path(row[ImRegBenchmark.COL_POINTS_REF],
pre_path=path_reference)))
ratio_matches = len(pairs) / float(len_lnds_ref)
return idx, ratio_matches
def compute_registration_statistic(cls,
idx_row,
df_experiments,
path_dataset=None,
path_experiment=None):
""" after successful registration load initial nad estimated landmarks
afterwords compute various statistic for init, and final alignment
:param tuple(int,dict) df_row: row from iterated table
:param DF df_experiments: DataFrame with experiments
:param str|None path_dataset: path to the dataset folder
:param str|None path_experiment: path to the experiment folder
"""
idx, row = idx_row
row = dict(row) # convert even series to dictionary
points_ref, points_move, path_img_ref = cls._load_landmarks(
row, path_dataset)
img_diag = cls._image_diag(row, path_img_ref)
df_experiments.loc[idx, cls.COL_IMAGE_DIAGONAL] = img_diag
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_ref,
points_move,
'init',
img_diag,
wo_affine=False)
# load transformed landmarks
if (cls.COL_POINTS_MOVE_WARP not in row) and (cls.COL_POINTS_REF_WARP
not in row):
logging.error('Statistic: no output landmarks')
return
# define what is the target and init state according to the experiment results
is_move_warp = isinstance(row.get(cls.COL_POINTS_MOVE_WARP, None), str)
points_init = points_move if is_move_warp else points_ref
points_target = points_ref if is_move_warp else points_move
col_lnds_warp = cls.COL_POINTS_MOVE_WARP if is_move_warp else cls.COL_POINTS_REF_WARP
# check if there are reference landmarks
if points_target is None:
logging.warning(
'Missing landmarks in "%s"',
cls.COL_POINTS_REF if is_move_warp else cls.COL_POINTS_MOVE)
return
# load warped landmarks
path_lnds_wapr = update_path(row[col_lnds_warp],
pre_path=path_experiment)
if path_lnds_wapr and os.path.isfile(path_lnds_wapr):
points_warp = load_landmarks(path_lnds_wapr)
points_warp = np.nan_to_num(points_warp)
else:
logging.warning('Invalid path to the landmarks: "%s" <- "%s"',
path_lnds_wapr, row[col_lnds_warp])
return
# compute Affine statistic
affine_diff = compute_affine_transf_diff(points_init, points_target,
points_warp)
for name in affine_diff:
df_experiments.loc[idx, name] = affine_diff[name]
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_target,
points_warp,
'elastic',
img_diag,
wo_affine=True)
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_target,
points_warp,
'target',
img_diag,
wo_affine=False)
row_ = dict(df_experiments.loc[idx])
# compute the robustness
if 'TRE Mean' in row_:
df_experiments.loc[idx, cls.COL_ROBUSTNESS] = \
compute_tre_robustness(points_target, points_init, points_warp)
def compute_registration_statistic(
cls,
idx_row,
df_experiments,
path_dataset=None,
path_experiment=None,
path_reference=None,
):
""" after successful registration load initial nad estimated landmarks
afterwords compute various statistic for init, and final alignment
:param tuple(int,dict) df_row: row from iterated table
:param DF df_experiments: DataFrame with experiments
:param str|None path_dataset: path to the provided dataset folder
:param str|None path_reference: path to the complete landmark collection folder
:param str|None path_experiment: path to the experiment folder
"""
idx, row = idx_row
row = dict(row) # convert even series to dictionary
# load common landmarks and image size
points_ref, points_move, path_img_ref = cls._load_landmarks(
row, path_dataset)
img_diag = cls._image_diag(row, path_img_ref)
df_experiments.loc[idx, cls.COL_IMAGE_DIAGONAL] = img_diag
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_ref,
points_move,
'init',
img_diag,
wo_affine=False)
# define what is the target and init state according to the experiment results
use_move_warp = isinstance(row.get(cls.COL_POINTS_MOVE_WARP), str)
if use_move_warp:
points_init, points_target = points_move, points_ref
col_source, col_target = cls.COL_POINTS_MOVE, cls.COL_POINTS_REF
col_lnds_warp = cls.COL_POINTS_MOVE_WARP
else:
points_init, points_target = points_ref, points_move
col_lnds_warp = cls.COL_POINTS_REF_WARP
col_source, col_target = cls.COL_POINTS_REF, cls.COL_POINTS_MOVE
# optional filtering
if path_reference:
ratio, points_target, _ = \
filter_paired_landmarks(row, path_dataset, path_reference, col_source, col_target)
df_experiments.loc[idx, COL_PAIRED_LANDMARKS] = np.round(ratio, 2)
# load transformed landmarks
if (cls.COL_POINTS_MOVE_WARP not in row) and (cls.COL_POINTS_REF_WARP
not in row):
logging.error('Statistic: no output landmarks')
return
# check if there are reference landmarks
if points_target is None:
logging.warning(
'Missing landmarks in "%s"',
cls.COL_POINTS_REF if use_move_warp else cls.COL_POINTS_MOVE)
return
# load warped landmarks
path_lnds_warp = update_path(row[col_lnds_warp],
pre_path=path_experiment)
if path_lnds_warp and os.path.isfile(path_lnds_warp):
points_warp = load_landmarks(path_lnds_warp)
points_warp = np.nan_to_num(points_warp)
else:
logging.warning('Invalid path to the landmarks: "%s" <- "%s"',
path_lnds_warp, row[col_lnds_warp])
return
df_experiments.loc[idx, cls.COL_NB_LANDMARKS_INPUT] = min(
len(points_init), len(points_target))
df_experiments.loc[idx, cls.COL_NB_LANDMARKS_WARP] = len(points_warp)
# compute Affine statistic
affine_diff = compute_affine_transf_diff(points_init, points_target,
points_warp)
for name in affine_diff:
df_experiments.loc[idx, name] = affine_diff[name]
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_target,
points_warp,
'elastic',
img_diag,
wo_affine=True)
# compute landmarks statistic
cls.compute_registration_accuracy(df_experiments,
idx,
points_target,
points_warp,
'target',
img_diag,
wo_affine=False)
row_ = dict(df_experiments.loc[idx])
# compute the robustness
if 'TRE Mean' in row_:
df_experiments.loc[idx, cls.COL_ROBUSTNESS] = \
compute_tre_robustness(points_target, points_init, points_warp)
def _load_landmarks(cls, item, path_dataset):
path_img_ref, _, path_lnds_ref, path_lnds_move = \
[update_path(item[col], pre_path=path_dataset) for col in cls.COVER_COLUMNS]
points_ref = load_landmarks(path_lnds_ref)
points_move = load_landmarks(path_lnds_move)
return points_ref, points_move, path_img_ref
import pandas as pd
from numpy.testing import assert_raises, assert_array_almost_equal
sys.path += [os.path.abspath('.'), os.path.abspath('..')] # Add path to root
from birl.utilities.data_io import update_path
from birl.utilities.dataset import args_expand_parse_images
from birl.utilities.experiments import (parse_arg_params, try_decorator)
from birl.cls_benchmark import ImRegBenchmark
from birl.cls_benchmark import (
NAME_CSV_RESULTS, NAME_TXT_RESULTS, NAME_CSV_REGISTRATION_PAIRS,
COVER_COLUMNS, COL_IMAGE_MOVE_WARP, COL_POINTS_REF_WARP,
COL_POINTS_MOVE_WARP, _visual_image_move_warp_lnds_move_warp,
_visual_image_ref_warp_lnds_move_warp, visualise_registration)
from birl.bm_template import BmTemplate
PATH_DATA = update_path('data_images')
PATH_CSV_COVER_MIX = os.path.join(PATH_DATA, 'pairs-imgs-lnds_mix.csv')
PATH_CSV_COVER_ANHIR = os.path.join(PATH_DATA, 'pairs-imgs-lnds_histol.csv')
# logging.basicConfig(level=logging.INFO)
class TestBmRegistration(unittest.TestCase):
@classmethod
def setUpClass(cls):
logging.basicConfig(level=logging.INFO)
path_base = os.path.dirname(update_path('requirements.txt'))
cls.path_out = os.path.join(path_base, 'output-test')
shutil.rmtree(cls.path_out, ignore_errors=True)
os.mkdir(cls.path_out)
def _remove_default_experiment(self, bm_name):
class BmUnwarpJ(ImRegBenchmark):
""" Benchmark for ImageJ plugin - bUnwarpJ
no run test while this method requires manual installation of ImageJ
For the app installation details, see module details.
EXAMPLE
-------
>>> from birl.utilities.data_io import create_folder, update_path
>>> path_out = create_folder('temp_results')
>>> fn_path_conf = lambda n: os.path.join(update_path('configs'), n)
>>> path_csv = os.path.join(update_path('data_images'), 'pairs-imgs-lnds_mix.csv')
>>> params = {'path_table': path_csv,
... 'path_out': path_out,
... 'exec_Fiji': 'ImageJ-linux64',
... 'preprocessing': ['hist-matching'],
... 'nb_workers': 2,
... 'unique': False,
... 'path_config': fn_path_conf('ImageJ_bUnwarpJ_histol.yaml')}
>>> benchmark = BmUnwarpJ(params)
>>> benchmark.run() # doctest: +SKIP
>>> params['path_config'] = fn_path_conf('ImageJ_bUnwarpJ-SIFT_histol.yaml')
>>> benchmark = BmUnwarpJ(params)
>>> benchmark.run() # doctest: +SKIP
>>> del benchmark
>>> shutil.rmtree(path_out, ignore_errors=True)
"""
#: required experiment parameters
REQUIRED_PARAMS = ImRegBenchmark.REQUIRED_PARAMS + [
'exec_Fiji', 'path_config'
]
#: path to IJ scripts
PATH_IJ_SCRIPTS = os.path.join(update_path('scripts'), 'ImageJ')
#: path/name of image registration script
PATH_SCRIPT_REGISTRATION_BASE = os.path.join(
PATH_IJ_SCRIPTS, 'apply-bUnwarpJ-registration.bsh')
#: path/name of image registration script with features
PATH_SCRIPT_REGISTRATION_SIFT = os.path.join(
PATH_IJ_SCRIPTS, 'apply-SIFT-bUnwarpJ-registration.bsh')
#: path/name of image/landmarks warping script
PATH_SCRIPT_WARP_LANDMARKS = os.path.join(PATH_IJ_SCRIPTS,
'apply-bUnwarpJ-transform.bsh')
# PATH_SCRIPT_HIST_MATCH_IJM = os.path.join(PATH_IJ_SCRIPTS,
# 'histogram-matching-for-macro.bsh')
#: command for executing the image registration
COMMAND_REGISTRATION = \
'%(exec_Fiji)s --headless %(path_bsh)s' \
' %(source)s %(target)s %(params)s' \
' %(output)s/transform-direct.txt' \
' %(output)s/transform-inverse.txt'
#: internal name of converted landmarks for tranf. script
NAME_LANDMARKS = 'source_landmarks.pts'
#: name of warped moving landmarks by tranf. script
NAME_LANDMARKS_WARPED = 'warped_source_landmarks.pts'
#: resulting inverse transformation
NAME_TRANSF_INVERSE = 'transform-inverse.txt'
#: resulting direct transformation
NAME_TRANSF_DIRECT = 'transform-direct.txt'
#: command for executing the warping image and landmarks
COMMAND_WARP_LANDMARKS = \
'%(exec_Fiji)s --headless %(path_bsh)s' \
' %(source)s %(target)s' \
' %(output)s/' + NAME_LANDMARKS + \
' %(output)s/' + NAME_LANDMARKS_WARPED + \
' %(transf-inv)s' \
' %(transf-dir)s' \
' %(warp)s'
#: required parameters in the configuration file for bUnwarpJ
REQUIRED_PARAMS_BUNWARPJ = ('mode', 'subsampleFactor', 'minScale',
'maxScale', 'divWeight', 'curlWeight',
'landmarkWeight', 'imageWeight',
'consistencyWeight', 'stopThreshold')
#: required parameters in the configuration file for SIFT features
REQUIRED_PARAMS_SIFT = ('initialSigma', 'steps', 'minOctaveSize',
'maxOctaveSize', 'fdSize', 'fdBins', 'rod',
'maxEpsilon', 'minInlierRatio', 'modelIndex')
#: default bUnwarpJ and SIFT parameters
DEFAULT_PARAMS = {
'bUnwarpJ': {
'mode': 1, #: (0-Accurate, 1-Fast, 2-Mono)
'subsampleFactor': 0, # (0 = 2^0, 7 = 2^7)
'minScale': 0, # (0-Very Coarse, 1-Coarse, 2-Fine, 3-Very Fine)
'maxScale':
3, # (0-Very Coarse, 1-Coarse, 2-Fine, 3-Very Fine, 4-Super Fine)
# weight to penalize divergence
'divWeight': 0.1,
#: weight to penalize curl
'curlWeight': 0.1,
#: weight to penalize landmark location error
'landmarkWeight': 0.,
#: weight to penalize intensity difference
'imageWeight': 1.,
#: weight to penalize consistency difference
'consistencyWeight': 10.,
#: error function stopping threshold value
'stopThreshold': 0.01,
},
'SIFT': {
# initial Gaussian blur sigma
'initialSigma': 1.6,
#: steps per scale octave
'steps': 3,
#: minimum image size in pixels
'minOctaveSize': 64,
#: maximum image size in pixels
'maxOctaveSize': 1024,
#: feature descriptor size
'fdSize': 8,
#: feature descriptor orientation bins
'fdBins': 8,
#: closest/next closest ratio
'rod': 0.92,
#: maximal alignment error in pixels
'maxEpsilon': 25,
#: inlier ratio
'minInlierRatio': 0.05,
#: expected transformation of range
'modelIndex':
1, # (0:Translation, 1:Rigid, 2:Similarity, 3:Affine, 4:Perspective)
}
}
# assert all(k in DEFAULT_PARAMS['bUnwarpJ'] for k in REQUIRED_PARAMS_BUNWARPJ), \
# 'default params are missing some required parameters for bUnwarpJ'
# assert all(k in DEFAULT_PARAMS['SIFT'] for k in REQUIRED_PARAMS_SIFT), \
# 'default params are missing some required parameters for SIFT'
def _prepare(self):
""" prepare Benchmark - copy configurations """
logging.info('-> copy configuration...')
self._copy_config_to_expt('path_config')
def _generate_regist_command(self, item):
""" generate the registration command(s)
:param dict item: dictionary with registration params
:return str|list(str): the execution commands
"""
path_im_ref, path_im_move, _, _ = self._get_paths(item,
prefer_pproc=True)
path_dir = self._get_path_reg_dir(item)
config = self.DEFAULT_PARAMS
config = dict_deep_update(config,
load_config_yaml(self.params['path_config']))
assert config['bUnwarpJ']['mode'] < 2, 'Mono mode does not supports inverse transform' \
' which is need for landmarks warping.'
config_sift = [config['SIFT'][k] for k in self.REQUIRED_PARAMS_SIFT] \
if config.get('SIFT', False) else []
config_bunwarpj = [
config['bUnwarpJ'][k] for k in self.REQUIRED_PARAMS_BUNWARPJ
]
path_reg_script = self.PATH_SCRIPT_REGISTRATION_SIFT if config_sift \
else self.PATH_SCRIPT_REGISTRATION_BASE
cmd = self.COMMAND_REGISTRATION % {
'exec_Fiji': self.params['exec_Fiji'],
'path_bsh': path_reg_script,
'target': path_im_ref,
'source': path_im_move,
'output': path_dir,
'params': ' '.join(map(str, config_sift + config_bunwarpj)),
}
return cmd
def _extract_warped_image_landmarks(self, item):
""" get registration results - warped registered images and landmarks
:param dict item: dictionary with registration params
:return dict: paths to warped images/landmarks
"""
logging.debug('.. warp the registered image and get landmarks')
path_dir = self._get_path_reg_dir(item)
path_im_ref, path_im_move, _, path_lnds_move = self._get_paths(
item, prefer_pproc=False)
path_log = os.path.join(path_dir, self.NAME_LOG_REGISTRATION)
# warp moving landmarks to reference frame
path_img_warp = os.path.join(path_dir, os.path.basename(path_im_move))
dict_params = {
'exec_Fiji': self.params['exec_Fiji'],
'path_bsh': self.PATH_SCRIPT_WARP_LANDMARKS,
'source': path_im_move,
'target': path_im_ref,
'output': path_dir,
'transf-inv': os.path.join(path_dir, self.NAME_TRANSF_INVERSE),
'transf-dir': os.path.join(path_dir, self.NAME_TRANSF_DIRECT),
'warp': path_img_warp,
}
# export source points to TXT
pts_source = load_landmarks(path_lnds_move)
save_landmarks(os.path.join(path_dir, self.NAME_LANDMARKS), pts_source)
# execute transformation
exec_commands(self.COMMAND_WARP_LANDMARKS % dict_params,
path_logger=path_log,
timeout=self.EXECUTE_TIMEOUT)
# load warped landmarks from TXT
path_lnds_warp = os.path.join(path_dir, self.NAME_LANDMARKS_WARPED)
if os.path.isfile(path_lnds_warp):
points_warp = load_landmarks(path_lnds_warp)
path_lnds_warp = os.path.join(path_dir,
os.path.basename(path_lnds_move))
save_landmarks(path_lnds_warp, points_warp)
else:
path_lnds_warp = None
# return results
return {
self.COL_IMAGE_MOVE_WARP: path_img_warp,
self.COL_POINTS_MOVE_WARP: path_lnds_warp
}
@staticmethod
def extend_parse(arg_parser):
""" extent the basic arg parses by some extra required parameters
:return object:
"""
# SEE: https://docs.python.org/3/library/argparse.html
arg_parser.add_argument('-Fiji',
'--exec_Fiji',
type=str,
required=True,
help='path to the Fiji executable')
arg_parser.add_argument('-cfg',
'--path_config',
required=True,
type=str,
help='path to the bUnwarpJ configuration')
return arg_parser
import os
import sys
import time
import logging
import json
import shutil
sys.path += [os.path.abspath('.'), os.path.abspath('..')] # Add path to root
from birl.utilities.data_io import update_path, load_landmarks, save_landmarks
from birl.utilities.experiments import create_basic_parse, parse_arg_params, exec_commands
from birl.cls_benchmark import ImRegBenchmark, NAME_LOG_REGISTRATION, COL_TIME
from birl.bm_template import main
from bm_experiments import bm_comp_perform
PATH_IJ_SCRIPTS = os.path.join(update_path('scripts'), 'ImageJ')
PATH_SCRIPT_REGISTRATION = os.path.join(PATH_IJ_SCRIPTS,
'apply-bUnwarpJ-registration.bsh')
PATH_SCRIPT_WARP_LANDMARKS = os.path.join(PATH_IJ_SCRIPTS,
'apply-bUnwarpJ-transform.bsh')
PATH_SCRIPT_HIST_MATCHING = os.path.join(PATH_IJ_SCRIPTS,
'histogram-matching.bsh')
# PATH_SCRIPT_HIST_MATCH_IJM = os.path.join(PATH_IJ_SCRIPTS, 'histogram-matching-for-macro.bsh')
NAME_LANDMARKS = 'source_landmarks.txt'
NAME_LANDMARKS_WARPED = 'warped_source_landmarks.txt'
COMMAND_WARP_LANDMARKS = '%(path_fiji)s --headless %(path_bsh)s' \
' %(source)s %(target)s' \
' %(output)s/' + NAME_LANDMARKS + \
' %(output)s/' + NAME_LANDMARKS_WARPED + \
' %(output)s/transform-inverse.txt' \
' %(output)s/transform-direct.txt' \
Check whether it generates correct outputs and resulting values
Copyright (C) 2017-2019 Jiri Borovec <*****@*****.**>
"""
import os
import sys
import unittest
from parameterized import parameterized
sys.path += [os.path.abspath('.'), os.path.abspath('..')] # Add path to root
from birl.utilities.data_io import update_path, load_image
from birl.utilities.dataset import image_histogram_matching, CONVERT_RGB
PATH_ROOT = os.path.dirname(update_path('birl'))
PATH_DATA = update_path('data_images')
PATH_IMAGE_REF = os.path.join(PATH_DATA, 'rat-kidney_', 'scale-5pc',
'Rat-Kidney_HE.jpg')
PATH_IMAGE_SRC = os.path.join(PATH_DATA, 'rat-kidney_', 'scale-5pc',
'Rat-Kidney_PanCytokeratin.jpg')
class TestHistogramMatching(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.img_ref = load_image(PATH_IMAGE_REF)
cls.img_src = load_image(PATH_IMAGE_SRC)
@parameterized.expand(list(CONVERT_RGB.keys()))
def test_hist_matching(self, clr_space):
def setUpClass(cls):
logging.basicConfig(level=logging.INFO)
path_base = os.path.dirname(update_path('requirements.txt'))
cls.path_out = os.path.join(path_base, 'output-test')
shutil.rmtree(cls.path_out, ignore_errors=True)
os.mkdir(cls.path_out)
