def convert_tiffs(new_tiff_filenames,
new_hdf5_pathname,
axis=0,
channel=0,
z_index=0,
pages_to_channel=1,
memmap=False):
"""
Convert a stack of tiffs to an HDF5 file.
Args:
new_tiff_filenames(list or str): takes a str for a single file
or a list of strs for
filenames to combine (allows
regex).
new_hdf5_pathname(str): the HDF5 file and location to
store the dataset.
axis(int): which axis to concatenate
along.
channel(int): which channel to select for the
HDF5 (can only keep one).
z_index(int): which z value to take (the
algorithm is not setup for 3D
data yet)
pages_to_channel(int): if channels are not normally
stored in the channel variable,
but are stored as pages, then
this will split neighboring
pages into separate channels.
memmap(bool): allows one to load the array
using a memory mapped file as
opposed to reading it directly.
(by default is False)
"""
assert (pages_to_channel > 0)
# Get the axes that do not change
static_axes = numpy.array(list(iters.xrange_with_skip(3, to_skip=axis)))
# if it is only a single str, make it a singleton list
if isinstance(new_tiff_filenames, (bytes, unicode)):
new_tiff_filenames = [new_tiff_filenames]
# Expand any regex in path names
new_tiff_filenames = xglob.expand_pathname_list(*new_tiff_filenames)
# Extract the offset and descriptions for storage.
new_hdf5_dataset_filenames = list()
new_hdf5_dataset_offsets = list()
# Determine the shape and dtype to use for the dataset (so that everything
# will fit).
new_hdf5_dataset_shape = numpy.zeros((3, ), dtype=int)
new_hdf5_dataset_dtype = bool
for each_new_tiff_filename in new_tiff_filenames:
# Add each filename.
new_hdf5_dataset_filenames.append(unicode(each_new_tiff_filename))
# Get all of the offsets.
new_hdf5_dataset_offsets.append(new_hdf5_dataset_shape[axis])
# Get the shape and type of each frame.
each_new_tiff_file_shape, each_new_tiff_file_dtype = list(
get_multipage_tiff_shape_dtype_transformed(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel).values())
each_new_tiff_file_shape = each_new_tiff_file_shape[2:]
# Find the increase on the merge axis. Find the largest shape for the
# rest.
each_new_tiff_file_shape = numpy.array(each_new_tiff_file_shape)
new_hdf5_dataset_shape[axis] += each_new_tiff_file_shape[axis]
new_hdf5_dataset_shape[static_axes] = numpy.array([
new_hdf5_dataset_shape[static_axes],
each_new_tiff_file_shape[static_axes]
]).max(axis=0)
# Finds the best type that everything can be cast to without loss of
# precision.
if not numpy.can_cast(each_new_tiff_file_dtype,
new_hdf5_dataset_dtype):
if numpy.can_cast(new_hdf5_dataset_dtype,
each_new_tiff_file_dtype):
new_hdf5_dataset_dtype = each_new_tiff_file_dtype
else:
raise Exception("Cannot find safe conversion between" +
" new_hdf5_dataset_dtype = " +
repr(new_hdf5_dataset_dtype) +
" and each_new_tiff_file_dtype = " +
repr(each_new_tiff_file_dtype) + ".")
# Convert to arrays.
new_hdf5_dataset_filenames = numpy.array(new_hdf5_dataset_filenames)
new_hdf5_dataset_offsets = numpy.array(new_hdf5_dataset_offsets)
# Convert to standard forms
new_hdf5_dataset_shape = tuple(new_hdf5_dataset_shape)
new_hdf5_dataset_dtype = numpy.dtype(new_hdf5_dataset_dtype)
# Get all the needed locations for the HDF5 file and dataset
new_hdf5_filename, new_hdf5_dataset_name = hdf5.serializers.split_hdf5_path(
new_hdf5_pathname)
new_hdf5_groupname = os.path.dirname(new_hdf5_dataset_name)
# Dump all datasets to the file
with h5py.File(new_hdf5_filename, "a") as new_hdf5_file:
new_hdf5_group = new_hdf5_file.require_group(new_hdf5_groupname)
new_hdf5_dataset = new_hdf5_group.create_dataset(
new_hdf5_dataset_name,
new_hdf5_dataset_shape,
new_hdf5_dataset_dtype,
chunks=True)
new_hdf5_dataset.attrs.create("filenames",
new_hdf5_dataset_filenames,
shape=new_hdf5_dataset_filenames.shape,
dtype=h5py.special_dtype(vlen=unicode))
new_hdf5_dataset.attrs["offsets"] = new_hdf5_dataset_offsets
# Workaround required due to this issue
# ( https://github.com/h5py/h5py/issues/289 ).
new_hdf5_descriptions_dataset = new_hdf5_group.create_dataset(
"_".join([new_hdf5_dataset_name, "descriptions"]),
shape=new_hdf5_dataset_shape[0:1],
dtype=h5py.special_dtype(vlen=unicode))
new_hdf5_dataset.attrs["descriptions"] = (
new_hdf5_descriptions_dataset.file.filename +
new_hdf5_descriptions_dataset.name)
new_hdf5_dataset_axis_pos = 0
for each_new_tiff_filename in new_tiff_filenames:
# Log the filename in case something goes wrong.
trace_logger.info("Now appending TIFF: \"" +
str(each_new_tiff_filename) + "\"")
# Read the data in the format specified.
each_new_tiff_array, each_new_tiff_description = get_standard_tiff_data(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel,
memmap=memmap)
# Take channel and z selection
# TODO: Could we drop the channel constraint?
# TODO: Want to drop z constraint.
each_new_tiff_array = each_new_tiff_array[channel, z_index]
each_new_tiff_description = each_new_tiff_description[channel]
# Store into the current slice and go to the next one.
new_hdf5_dataset_axis_pos_next = new_hdf5_dataset_axis_pos + \
len(each_new_tiff_array)
new_hdf5_dataset[
new_hdf5_dataset_axis_pos:
new_hdf5_dataset_axis_pos_next] = each_new_tiff_array
new_hdf5_descriptions_dataset[
new_hdf5_dataset_axis_pos:
new_hdf5_dataset_axis_pos_next] = each_new_tiff_description
new_hdf5_dataset_axis_pos = new_hdf5_dataset_axis_pos_next
def convert_tiffs(new_tiff_filenames,
new_hdf5_pathname,
axis=0,
channel=0,
z_index=0,
pages_to_channel=1,
memmap=False):
"""
Convert a stack of tiffs to an HDF5 file.
Args:
new_tiff_filenames(list or str): takes a str for a single file
or a list of strs for
filenames to combine (allows
regex).
new_hdf5_pathname(str): the HDF5 file and location to
store the dataset.
axis(int): which axis to concatenate
along.
channel(int): which channel to select for the
HDF5 (can only keep one).
z_index(int): which z value to take (the
algorithm is not setup for 3D
data yet)
pages_to_channel(int): if channels are not normally
stored in the channel variable,
but are stored as pages, then
this will split neighboring
pages into separate channels.
memmap(bool): allows one to load the array
using a memory mapped file as
opposed to reading it directly.
(by default is False)
"""
assert (pages_to_channel > 0)
# Get the axes that do not change
static_axes = numpy.array(list(iters.xrange_with_skip(
3, to_skip=axis
)))
# if it is only a single str, make it a singleton list
if isinstance(new_tiff_filenames, (bytes, unicode)):
new_tiff_filenames = [new_tiff_filenames]
# Expand any regex in path names
new_tiff_filenames = xglob.expand_pathname_list(*new_tiff_filenames)
# Extract the offset and descriptions for storage.
new_hdf5_dataset_filenames = list()
new_hdf5_dataset_offsets = list()
# Determine the shape and dtype to use for the dataset (so that everything
# will fit).
new_hdf5_dataset_shape = numpy.zeros((3,), dtype=int)
new_hdf5_dataset_dtype = bool
for each_new_tiff_filename in new_tiff_filenames:
# Add each filename.
new_hdf5_dataset_filenames.append(unicode(each_new_tiff_filename))
# Get all of the offsets.
new_hdf5_dataset_offsets.append(new_hdf5_dataset_shape[axis])
# Get the shape and type of each frame.
each_new_tiff_file_shape, each_new_tiff_file_dtype = list(get_multipage_tiff_shape_dtype_transformed(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel
).values())
each_new_tiff_file_shape = each_new_tiff_file_shape[2:]
# Find the increase on the merge axis. Find the largest shape for the
# rest.
each_new_tiff_file_shape = numpy.array(each_new_tiff_file_shape)
new_hdf5_dataset_shape[axis] += each_new_tiff_file_shape[axis]
new_hdf5_dataset_shape[static_axes] = numpy.array(
[
new_hdf5_dataset_shape[static_axes],
each_new_tiff_file_shape[static_axes]
]
).max(axis=0)
# Finds the best type that everything can be cast to without loss of
# precision.
if not numpy.can_cast(each_new_tiff_file_dtype, new_hdf5_dataset_dtype):
if numpy.can_cast(new_hdf5_dataset_dtype, each_new_tiff_file_dtype):
new_hdf5_dataset_dtype = each_new_tiff_file_dtype
else:
raise Exception(
"Cannot find safe conversion between" +
" new_hdf5_dataset_dtype = " +
repr(new_hdf5_dataset_dtype) +
" and each_new_tiff_file_dtype = " +
repr(each_new_tiff_file_dtype) + "."
)
# Convert to arrays.
new_hdf5_dataset_filenames = numpy.array(new_hdf5_dataset_filenames)
new_hdf5_dataset_offsets = numpy.array(new_hdf5_dataset_offsets)
# Convert to standard forms
new_hdf5_dataset_shape = tuple(new_hdf5_dataset_shape)
new_hdf5_dataset_dtype = numpy.dtype(new_hdf5_dataset_dtype)
# Get all the needed locations for the HDF5 file and dataset
new_hdf5_filename, new_hdf5_dataset_name = hdf5.serializers.split_hdf5_path(new_hdf5_pathname)
new_hdf5_groupname = os.path.dirname(new_hdf5_dataset_name)
# Dump all datasets to the file
with h5py.File(new_hdf5_filename, "a") as new_hdf5_file:
new_hdf5_group = new_hdf5_file.require_group(new_hdf5_groupname)
new_hdf5_dataset = new_hdf5_group.create_dataset(
new_hdf5_dataset_name,
new_hdf5_dataset_shape,
new_hdf5_dataset_dtype,
chunks=True
)
new_hdf5_dataset.attrs.create(
"filenames",
new_hdf5_dataset_filenames,
shape=new_hdf5_dataset_filenames.shape,
dtype=h5py.special_dtype(vlen=unicode)
)
new_hdf5_dataset.attrs["offsets"] = new_hdf5_dataset_offsets
# Workaround required due to this issue
# ( https://github.com/h5py/h5py/issues/289 ).
new_hdf5_descriptions_dataset = new_hdf5_group.create_dataset(
"_".join([new_hdf5_dataset_name, "descriptions"]),
shape=new_hdf5_dataset_shape[0:1],
dtype=h5py.special_dtype(vlen=unicode)
)
new_hdf5_dataset.attrs["descriptions"] = (
new_hdf5_descriptions_dataset.file.filename +
new_hdf5_descriptions_dataset.name
)
new_hdf5_dataset_axis_pos = 0
for each_new_tiff_filename in new_tiff_filenames:
# Log the filename in case something goes wrong.
trace_logger.info(
"Now appending TIFF: \"" + str(each_new_tiff_filename) + "\""
)
# Read the data in the format specified.
each_new_tiff_array, each_new_tiff_description = get_standard_tiff_data(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel,
memmap=memmap
)
# Take channel and z selection
# TODO: Could we drop the channel constraint?
# TODO: Want to drop z constraint.
each_new_tiff_array = each_new_tiff_array[channel, z_index]
each_new_tiff_description = each_new_tiff_description[channel]
# Store into the current slice and go to the next one.
new_hdf5_dataset_axis_pos_next = new_hdf5_dataset_axis_pos + \
len(each_new_tiff_array)
new_hdf5_dataset[new_hdf5_dataset_axis_pos:new_hdf5_dataset_axis_pos_next] = each_new_tiff_array
new_hdf5_descriptions_dataset[new_hdf5_dataset_axis_pos:new_hdf5_dataset_axis_pos_next] = each_new_tiff_description
new_hdf5_dataset_axis_pos = new_hdf5_dataset_axis_pos_next
def convert_tiffs(new_tiff_filenames, new_hdf5_pathname, axis = 0, channel = 0, z_index = 0, pages_to_channel = 1):
"""
Convert a stack of tiffs to an HDF5 file.
Args:
new_tiff_filenames(list or str): takes a str for a single file or a list of strs for filenames to combine
(allows regex).
new_hdf5_pathname(str): the HDF5 file and location to store the dataset.
axis(int): which axis to concatenate along.
channel(int): which channel to select for the HDF5 (can only keep one).
z_index(int): which z value to take (the algorithm is not setup for 3D data yet)
pages_to_channel(int): if channels are not normally stored in the channel variable, but are
stored as pages, then this will split neighboring pages into separate
channels.
"""
assert (pages_to_channel > 0)
# Get the axes that do not change
static_axes = numpy.array(list(iters.xrange_with_skip(3, to_skip = axis)))
# if it is only a single str, make it a singleton list
if isinstance(new_tiff_filenames, str):
new_tiff_filenames = [new_tiff_filenames]
# Expand any regex in path names
new_tiff_filenames = xglob.expand_pathname_list(*new_tiff_filenames)
# Determine the shape and dtype to use for the dataset (so that everything will fit).
new_hdf5_dataset_shape = numpy.zeros((3,), dtype = int)
new_hdf5_dataset_dtype = bool
for each_new_tiff_filename in new_tiff_filenames:
each_new_tiff_file_shape, each_new_tiff_file_dtype = get_multipage_tiff_shape_dtype_transformed(each_new_tiff_filename,
axis_order = "cztyx",
pages_to_channel = pages_to_channel).values()
each_new_tiff_file_shape = each_new_tiff_file_shape[2:]
# Find the increase on the merge axis. Find the largest shape for the rest.
each_new_tiff_file_shape = numpy.array(each_new_tiff_file_shape)
new_hdf5_dataset_shape[axis] += each_new_tiff_file_shape[axis]
new_hdf5_dataset_shape[static_axes] = numpy.array([new_hdf5_dataset_shape[static_axes], each_new_tiff_file_shape[static_axes]]).max(axis=0)
# Finds the best type that everything can be cast to without loss of precision.
if not numpy.can_cast(each_new_tiff_file_dtype, new_hdf5_dataset_dtype):
if numpy.can_cast(new_hdf5_dataset_dtype, each_new_tiff_file_dtype):
new_hdf5_dataset_dtype = each_new_tiff_file_dtype
else:
raise Exception("Cannot find safe conversion between new_hdf5_dataset_dtype = " + repr(new_hdf5_dataset_dtype) + " and each_new_tiff_file_dtype = " + repr(each_new_tiff_file_dtype) + ".")
# Convert to standard forms
new_hdf5_dataset_shape = tuple(new_hdf5_dataset_shape)
new_hdf5_dataset_dtype = numpy.dtype(new_hdf5_dataset_dtype)
# Get all the needed locations for the HDF5 file and dataset
new_hdf5_path_components = pathHelpers.PathComponents(new_hdf5_pathname)
new_hdf5_filename = new_hdf5_path_components.externalPath
new_hdf5_groupname = new_hdf5_path_components.internalDirectory
new_hdf5_dataset_name = new_hdf5_path_components.internalPath
# Dump all datasets to the file
with h5py.File(new_hdf5_filename, "a") as new_hdf5_file:
if new_hdf5_groupname not in new_hdf5_file:
new_hdf5_file.create_group(new_hdf5_groupname)
new_hdf5_group = new_hdf5_file[new_hdf5_groupname]
new_hdf5_dataset = new_hdf5_group.create_dataset(new_hdf5_dataset_name,
new_hdf5_dataset_shape,
new_hdf5_dataset_dtype,
chunks=True)
new_hdf5_dataset_axis_pos = 0
for each_new_tiff_filename in new_tiff_filenames:
# Read the data in the format specified.
each_new_tiff_array = get_standard_tiff_array(each_new_tiff_filename,
axis_order = "cztyx",
pages_to_channel = pages_to_channel)
# Take channel and z selection
# TODO: Could we drop the channel constraint by saving different channels to different arrays? Need to think about it.
# TODO: Want to drop z constraint, but need to consult with Ferran about algorithms that work on 3D for the end.
each_new_tiff_array = each_new_tiff_array[channel, z_index]
# Store into the current slice and go to the next one.
new_hdf5_dataset_axis_pos_next = new_hdf5_dataset_axis_pos + len(each_new_tiff_array)
new_hdf5_dataset[ new_hdf5_dataset_axis_pos : new_hdf5_dataset_axis_pos_next ] = each_new_tiff_array
new_hdf5_dataset_axis_pos = new_hdf5_dataset_axis_pos_next
def convert_tiffs(new_tiff_filenames,
new_hdf5_pathname,
axis=0,
channel=0,
z_index=0,
pages_to_channel=1):
"""
Convert a stack of tiffs to an HDF5 file.
Args:
new_tiff_filenames(list or str): takes a str for a single file
or a list of strs for
filenames to combine (allows
regex).
new_hdf5_pathname(str): the HDF5 file and location to
store the dataset.
axis(int): which axis to concatenate
along.
channel(int): which channel to select for the
HDF5 (can only keep one).
z_index(int): which z value to take (the
algorithm is not setup for 3D
data yet)
pages_to_channel(int): if channels are not normally
stored in the channel variable,
but are stored as pages, then
this will split neighboring
pages into separate channels.
"""
assert (pages_to_channel > 0)
# Get the axes that do not change
static_axes = numpy.array(list(iters.xrange_with_skip(3, to_skip=axis)))
# if it is only a single str, make it a singleton list
if isinstance(new_tiff_filenames, str):
new_tiff_filenames = [new_tiff_filenames]
# Expand any regex in path names
new_tiff_filenames = xglob.expand_pathname_list(*new_tiff_filenames)
# Extract the offset and descriptions for storage.
new_hdf5_dataset_filenames = list()
new_hdf5_dataset_offsets = list()
# Determine the shape and dtype to use for the dataset (so that everything
# will fit).
new_hdf5_dataset_shape = numpy.zeros((3,), dtype=int)
new_hdf5_dataset_dtype = bool
for each_new_tiff_filename in new_tiff_filenames:
# Add each filename.
new_hdf5_dataset_filenames.append(each_new_tiff_filename)
# Get all of the offsets.
new_hdf5_dataset_offsets.append(new_hdf5_dataset_shape[axis])
# Get the shape and type of each frame.
each_new_tiff_file_shape, each_new_tiff_file_dtype = get_multipage_tiff_shape_dtype_transformed(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel
).values()
each_new_tiff_file_shape = each_new_tiff_file_shape[2:]
# Find the increase on the merge axis. Find the largest shape for the
# rest.
each_new_tiff_file_shape = numpy.array(each_new_tiff_file_shape)
new_hdf5_dataset_shape[axis] += each_new_tiff_file_shape[axis]
new_hdf5_dataset_shape[static_axes] = numpy.array(
[
new_hdf5_dataset_shape[static_axes],
each_new_tiff_file_shape[static_axes]
]
).max(axis=0)
# Finds the best type that everything can be cast to without loss of
# precision.
if not numpy.can_cast(each_new_tiff_file_dtype, new_hdf5_dataset_dtype):
if numpy.can_cast(new_hdf5_dataset_dtype, each_new_tiff_file_dtype):
new_hdf5_dataset_dtype = each_new_tiff_file_dtype
else:
raise Exception(
"Cannot find safe conversion between" +
" new_hdf5_dataset_dtype = " +
repr(new_hdf5_dataset_dtype) +
" and each_new_tiff_file_dtype = " +
repr(each_new_tiff_file_dtype) + "."
)
# Convert to arrays.
new_hdf5_dataset_filenames = numpy.array(new_hdf5_dataset_filenames)
new_hdf5_dataset_offsets = numpy.array(new_hdf5_dataset_offsets)
# Convert to standard forms
new_hdf5_dataset_shape = tuple(new_hdf5_dataset_shape)
new_hdf5_dataset_dtype = numpy.dtype(new_hdf5_dataset_dtype)
# Get all the needed locations for the HDF5 file and dataset
new_hdf5_path_components = pathHelpers.PathComponents(new_hdf5_pathname)
new_hdf5_filename = new_hdf5_path_components.externalPath
new_hdf5_groupname = new_hdf5_path_components.internalDirectory
new_hdf5_dataset_name = new_hdf5_path_components.internalPath
# Dump all datasets to the file
with h5py.File(new_hdf5_filename, "a") as new_hdf5_file:
new_hdf5_group = new_hdf5_file.require_group(new_hdf5_groupname)
new_hdf5_dataset = new_hdf5_group.create_dataset(
new_hdf5_dataset_name,
new_hdf5_dataset_shape,
new_hdf5_dataset_dtype,
chunks=True
)
new_hdf5_dataset.attrs["filenames"] = new_hdf5_dataset_filenames
new_hdf5_dataset.attrs["offsets"] = new_hdf5_dataset_offsets
# Workaround required due to this issue
# ( https://github.com/h5py/h5py/issues/289 ).
new_hdf5_descriptions_dataset = new_hdf5_group.create_dataset(
"_".join([new_hdf5_dataset_name, "descriptions"]),
shape=new_hdf5_dataset_shape[0:1],
dtype=h5py.special_dtype(vlen=unicode)
)
new_hdf5_dataset.attrs["descriptions"] = (
new_hdf5_descriptions_dataset.file.filename +
new_hdf5_descriptions_dataset.name
)
new_hdf5_dataset_axis_pos = 0
for each_new_tiff_filename in new_tiff_filenames:
# Read the data in the format specified.
each_new_tiff_array = get_standard_tiff_array(
each_new_tiff_filename,
axis_order="cztyx",
pages_to_channel=pages_to_channel
)
# Extract the descriptions.
each_new_tiff_description = []
each_new_tiff_file = None
try:
each_new_tiff_file = libtiff.TiffFile(
each_new_tiff_filename, 'r'
)
for i in xrange(each_new_tiff_file.get_depth()):
metadata_i = each_new_tiff_file.IFD[i].entries_dict
desc_i = u""
try:
desc_i = unicode(
metadata_i["ImageDescription"].human()
)
except KeyError:
pass
each_new_tiff_description.append(
desc_i
)
finally:
if each_new_tiff_file:
each_new_tiff_file.close()
each_new_tiff_file = None
each_new_tiff_description = numpy.array(each_new_tiff_description)
# Take channel and z selection
# TODO: Could we drop the channel constraint?
# TODO: Want to drop z constraint.
each_new_tiff_array = each_new_tiff_array[channel, z_index]
# Store into the current slice and go to the next one.
new_hdf5_dataset_axis_pos_next = new_hdf5_dataset_axis_pos + \
len(each_new_tiff_array)
new_hdf5_dataset[new_hdf5_dataset_axis_pos:new_hdf5_dataset_axis_pos_next] = each_new_tiff_array
new_hdf5_descriptions_dataset[new_hdf5_dataset_axis_pos:new_hdf5_dataset_axis_pos_next] = each_new_tiff_description
new_hdf5_dataset_axis_pos = new_hdf5_dataset_axis_pos_next
