def _wrapper(func):
"""Wrapper that registers a function if it satisfies requirements."""
try:
auto_d = auto_discover
require_module(requires, min_version=min_version, mandatory=True)
AVAILABLE_LOADERS[name] = dict(func=func, auto_discover=auto_d)
except (ImportError, AssertionError):
pass
SUPPORTED_LOADERS[name] = (requires, min_version)
return func
def _wrapper(func):
"""Wrapper that registers a function if it satisfies requirements."""
try:
auto_d = auto_discover
require_module(requires, min_version=min_version, mandatory=True)
AVAILABLE_LOADERS[name] = dict(func=func, auto_discover=auto_d)
except (ImportError, AssertionError):
pass
SUPPORTED_LOADERS[name] = (requires, min_version)
return func
def _normalise_image(self, image):
"""
Normalises a 2D RGB image, if necessary performs any type casting
and reshaping operations.
:param image: a 2D RGB image, possibly given as a 5D tensor
:return: the normalised image in its original shape
"""
if isinstance(image.dtype, np.floating) and image.dtype != np.float32:
image = image.astype(np.float32)
elif isinstance(image.dtype, np.uint):
image = image.astype(np.float32)/255
orig_shape = list(image.shape)
if len(orig_shape) == 5 and (orig_shape[2] > 1 or orig_shape[3] > 1):
raise ValueError('Can only process 2D images.')
if len(image.shape) != 3:
image = image.reshape(orig_shape[:2] + [orig_shape[-1]])
image = image[...,::-1]
cv2 = require_module('cv2')
yuv_image = cv2.cvtColor(image, cv2.COLOR_BGR2YUV)
intensity = (255*yuv_image[...,0]).astype(np.uint8)
yuv_image[...,0] = cv2.equalizeHist(intensity).astype(np.float32)/255
return cv2.cvtColor(yuv_image, cv2.COLOR_YUV2BGR)[...,::-1]\
.reshape(orig_shape)
def imread_opencv(filename):
"""OpenCV image loader with identity 2D affine."""
cv2 = require_module('cv2')
img = cv2.imread(filename, flags=-1)
if img is None:
raise IOError(filename)
return image2nibabel(img[..., ::-1])
def imread_opencv(filename):
"""OpenCV image loader with identity 2D affine."""
cv2 = require_module('cv2')
img = cv2.imread(filename, flags=-1)
if img is None:
raise IOError(filename)
return image2nibabel(img[..., ::-1])
def _image3_animated_gif(tag, ims):
PIL = require_module('PIL')
from PIL.GifImagePlugin import Image as GIF
# x=numpy.random.randint(0,256,[10,10,10],numpy.uint8)
ims = [
np.asarray((ims[i, :, :]).astype(np.uint8))
for i in range(ims.shape[0])
]
ims = [GIF.fromarray(im) for im in ims]
img_str = b''
for b_data in PIL.GifImagePlugin.getheader(ims[0])[0]:
img_str += b_data
img_str += b'\x21\xFF\x0B\x4E\x45\x54\x53\x43\x41\x50' \
b'\x45\x32\x2E\x30\x03\x01\x00\x00\x00'
for i in ims:
for b_data in PIL.GifImagePlugin.getdata(i):
img_str += b_data
img_str += b'\x3B'
if IS_PYTHON2:
img_str = str(img_str)
summary_image_str = summary_pb2.Summary.Image(
height=10, width=10, colorspace=1, encoded_image_string=img_str)
image_summary = summary_pb2.Summary.Value(tag=tag, image=summary_image_str)
return [summary_pb2.Summary(value=[image_summary]).SerializeToString()]
def imread_sitk(filename):
"""SimpleITK requires two function calls to retrieve a numpy array."""
sitk = require_module('SimpleITK')
try:
simg = sitk.ReadImage(filename)
except RuntimeError:
raise IOError(filename)
img = sitk.GetArrayFromImage(simg)
return image2nibabel(img, affine=make_affine_from_sitk(simg))
def imread_sitk(filename):
"""SimpleITK requires two function calls to retrieve a numpy array."""
sitk = require_module('SimpleITK')
try:
simg = sitk.ReadImage(filename)
except RuntimeError:
raise IOError(filename)
img = sitk.GetArrayFromImage(simg)
if simg.GetDimension() > 2:
img = img.transpose()
return image2nibabel(img, affine=make_affine_from_sitk(simg))
def test_equilisation(self):
cv2 = require_module('cv2', mandatory=False)
if cv2 is None:
self.skipTest('requires cv2 module')
return
def _get_histogram(img):
inten = cv2.cvtColor(img[::-1], cv2.COLOR_BGR2YUV)[..., 0] * 255
return np.histogram(inten, 32, [0, 256])[0]
hist_before = _get_histogram(IMAGE_DATA)
layer = RGBHistogramEquilisationLayer(image_name='image')
orig_shape = list(IMAGE_DATA.shape)
input_shape = orig_shape[:2] + [1] * 2 + [3]
img, _ = layer(IMAGE_DATA.reshape(input_shape))
hist_after = _get_histogram(img.reshape(orig_shape))
self.assertGreater(
hist_before.astype(np.float32).std(),
hist_after.astype(np.float32).std())
img, _ = layer({'image': IMAGE_DATA.reshape(input_shape)})
hist_after = _get_histogram(img['image'].reshape(orig_shape))
self.assertGreater(
hist_before.astype(np.float32).std(),
hist_after.astype(np.float32).std())
img = (255 * IMAGE_DATA).astype(np.uint8)
img, _ = layer({'image': IMAGE_DATA.reshape(input_shape)})
hist_after = _get_histogram(img['image'].reshape(orig_shape))
self.assertGreater(
hist_before.astype(np.float32).std(),
hist_after.astype(np.float32).std())
def _image3_animated_gif(tag, ims):
PIL = require_module('PIL')
from PIL.GifImagePlugin import Image as GIF
# x=numpy.random.randint(0,256,[10,10,10],numpy.uint8)
ims = [np.asarray((ims[i, :, :]).astype(np.uint8))
for i in range(ims.shape[0])]
ims = [GIF.fromarray(im) for im in ims]
s = b''
for b in PIL.GifImagePlugin.getheader(ims[0])[0]:
s += b
s += b'\x21\xFF\x0B\x4E\x45\x54\x53\x43\x41\x50' \
b'\x45\x32\x2E\x30\x03\x01\x00\x00\x00'
for i in ims:
for b in PIL.GifImagePlugin.getdata(i):
s += b
s += b'\x3B'
if IS_PYTHON2:
s = str(s)
summary_image_str = summary_pb2.Summary.Image(
height=10, width=10, colorspace=1, encoded_image_string=s)
image_summary = summary_pb2.Summary.Value(
tag=tag, image=summary_image_str)
return [summary_pb2.Summary(value=[image_summary]).SerializeToString()]
def test_no_package(self):
with self.assertRaisesRegexp(ImportError, ''):
require_module('foobar_wrong_case')
def test_installed_min_version(self):
require_module('tensorflow', 1.0)
def imread_pillow(filename):
"""PIL (Pillow) image loader with an identity affine matrix."""
pil = require_module('PIL.Image')
img = np.asarray(pil.open(filename))
return image2nibabel(img)
def test_installed(self):
require_module('tensorflow')
def add_network_args(parser):
"""
keywords defined for network specification
:param parser:
:return:
"""
import niftynet.layer.binary_masking
import niftynet.layer.activation
import niftynet.utilities.histogram_standardisation as hist_std_module
parser.add_argument(
"--name",
help="Choose a net from NiftyNet/niftynet/network/ or from "
"user specified module string",
metavar='')
parser.add_argument(
"--activation_function",
help="Specify activation function types",
choices=list(niftynet.layer.activation.SUPPORTED_OP),
metavar='TYPE_STR',
default='relu')
parser.add_argument(
"--batch_size",
metavar='',
help="Set batch size of the net",
type=int,
default=2)
parser.add_argument(
"--decay",
help="[Training only] Set weight decay",
type=float,
default=0.0)
parser.add_argument(
"--reg_type",
metavar='TYPE_STR',
help="[Training only] Specify regulariser type_str",
type=str,
default='L2')
parser.add_argument(
"--volume_padding_size",
metavar='',
help="Set padding size of each volume (in all dimensions)",
type=spatialnumarray,
default=(0, 0, 0))
parser.add_argument(
"--window_sampling",
metavar='TYPE_STR',
help="How to sample patches from each loaded image:"
" 'uniform': fixed size uniformly distributed,"
" 'resize': resize image to the patch size.",
choices=['uniform', 'resize', 'balanced', 'weighted'],
default='uniform')
parser.add_argument(
"--queue_length",
help="Set size of preprocessing buffer queue",
metavar='',
type=int,
default=5)
parser.add_argument(
"--multimod_foreground_type",
choices=list(
niftynet.layer.binary_masking.SUPPORTED_MULTIMOD_MASK_TYPES),
help="Way of combining the foreground masks from different "
"modalities. 'and' is the intersection, 'or' is the union "
"and 'multi' permits each modality to use its own mask.",
default='and')
parser.add_argument(
"--histogram_ref_file",
metavar='',
type=str,
help="A reference file of histogram for intensity normalisation",
default=DEFAULT_HISTOGRAM_REF_FILE)
parser.add_argument(
"--norm_type",
help="Type of normalisation to perform",
type=str,
default='percentile',
choices=list(hist_std_module.SUPPORTED_CUTPOINTS))
parser.add_argument(
"--cutoff",
help="Cutoff values for the normalisation process",
type=float_array,
default=(0.01, 0.99))
parser.add_argument(
"--foreground_type",
choices=list(
niftynet.layer.binary_masking.SUPPORTED_MASK_TYPES),
help="type_str of foreground masking strategy used",
default='otsu_plus')
parser.add_argument(
"--normalisation",
help="Indicates if the normalisation must be performed",
type=str2boolean,
default=False)
parser.add_argument(
"--whitening",
help="Indicates if the whitening of the data should be applied",
type=str2boolean,
default=False)
parser.add_argument(
"--normalise_foreground_only",
help="Indicates whether a foreground mask should be applied when"
" normalising volumes",
type=str2boolean,
default=False)
parser.add_argument(
"--weight_initializer",
help="Set the initializer for the weight parameters",
type=str,
default='he_normal')
parser.add_argument(
"--bias_initializer",
help="Set the initializer for the bias parameters",
type=str,
default='zeros')
yaml = require_module('yaml', mandatory=False)
if yaml:
parser.add_argument(
"--weight_initializer_args",
help="Pass arguments to the initializer for the weight parameters",
type=yaml.load,
default={})
parser.add_argument(
"--bias_initializer_args",
help="Pass arguments to the initializer for the bias parameters",
type=yaml.load,
default={})
return parser
def test_no_version_info(self):
require_module('importlib', 0)
def test_no_version_info(self):
require_module('importlib', 0)
def test_wrong_version(self):
with self.assertRaisesRegexp(AssertionError, ''):
require_module('tensorflow', 100, mandatory=True)
def test_self_version(self):
require_module('importlib')
def test_no_package(self):
with self.assertRaisesRegexp(ImportError, ''):
require_module('foobar_wrong_case', mandatory=True)
def test_installed_min_version(self):
require_module('tensorflow', 1.0)
Data augmentation using elastic deformations as used by:
Milletari,F., Navab, N., & Ahmadi, S. A. (2016) V-net:
Fully convolutional neural networks for volumetric medical
image segmentation
"""
from __future__ import absolute_import, print_function
import warnings
import numpy as np
from niftynet.layer.base_layer import RandomisedLayer
from niftynet.utilities.util_import import require_module
sitk = require_module('SimpleITK')
warnings.simplefilter("ignore", UserWarning)
warnings.simplefilter("ignore", RuntimeWarning)
class RandomElasticDeformationLayer(RandomisedLayer):
"""
generate randomised elastic deformations
along each dim for data augmentation
"""
def __init__(self,
num_controlpoints=4,
std_deformation_sigma=15,
proportion_to_augment=0.5,
def test_wrong_version(self):
with self.assertRaisesRegexp(AssertionError, ''):
require_module('tensorflow', 100)
def add_network_args(parser):
"""
keywords defined for network specification
:param parser:
:return:
"""
import niftynet.layer.binary_masking
import niftynet.layer.activation
import niftynet.utilities.histogram_standardisation as hist_std_module
parser.add_argument(
"--name",
help="Choose a net from NiftyNet/niftynet/network/ or from "
"user specified module string",
metavar='')
parser.add_argument("--activation_function",
help="Specify activation function types",
choices=list(niftynet.layer.activation.SUPPORTED_OP),
metavar='TYPE_STR',
default='relu')
parser.add_argument("--batch_size",
metavar='',
help="Set batch size of the net",
type=int,
default=2)
parser.add_argument("--smaller_final_batch_mode",
metavar='TYPE_STR',
help="If True, allow the final batch to be smaller "
"if there are insufficient items left in the queue, "
"and the batch size will be undetermined during "
"graph construction.",
choices=list(SMALLER_FINAL_BATCH_MODE),
default='pad')
parser.add_argument("--decay",
help="[Training only] Set weight decay",
type=float,
default=0.0)
parser.add_argument("--reg_type",
metavar='TYPE_STR',
help="[Training only] Specify regulariser type_str",
type=str,
default='L2')
parser.add_argument(
"--volume_padding_size",
metavar='',
help="Set padding size of each volume (in all dimensions)",
type=spatialnumarray,
default=(0, 0, 0))
parser.add_argument("--volume_padding_mode",
metavar='',
help="Set which type of numpy padding to do, see "
"https://docs.scipy.org/doc/numpy-1.14.0/"
"reference/generated/numpy.pad.html "
"for details",
type=str,
default='minimum')
parser.add_argument(
"--volume_padding_to_size",
help="Choose size to pad all input volumes to. Any dimensions "
"that exceed the desired size will be kept the same. Default: "
"(0, ) which indicates not to use this mode. ",
type=spatialnumarray,
default=(0, ))
parser.add_argument(
"--window_sampling",
metavar='TYPE_STR',
help="How to sample patches from each loaded image:"
" 'uniform': fixed size uniformly distributed,"
" 'resize': resize image to the patch size.",
choices=['uniform', 'resize', 'balanced', 'weighted', 'patch'],
default='uniform')
parser.add_argument(
"--force_output_identity_resizing",
metavar=str2boolean,
help="Forces the shape of the inferred output to match the "
"input label shape rather than be resized to input image shape.",
default=False)
parser.add_argument("--queue_length",
help="Set size of preprocessing buffer queue",
metavar='',
type=int,
default=5)
parser.add_argument(
"--multimod_foreground_type",
choices=list(
niftynet.layer.binary_masking.SUPPORTED_MULTIMOD_MASK_TYPES),
help="Way of combining the foreground masks from different "
"modalities. 'and' is the intersection, 'or' is the union "
"and 'multi' permits each modality to use its own mask.",
default='and')
parser.add_argument(
"--histogram_ref_file",
metavar='',
type=str,
help="A reference file of histogram for intensity normalisation",
default=DEFAULT_HISTOGRAM_REF_FILE)
parser.add_argument("--norm_type",
help="Type of normalisation to perform",
type=str,
default='percentile',
choices=list(hist_std_module.SUPPORTED_CUTPOINTS))
parser.add_argument("--cutoff",
help="Cutoff values for the normalisation process",
type=float_array,
default=(0.01, 0.99))
parser.add_argument(
"--foreground_type",
choices=list(niftynet.layer.binary_masking.SUPPORTED_MASK_TYPES),
help="type_str of foreground masking strategy used",
default='otsu_plus')
parser.add_argument(
"--normalisation",
help="Indicates if the normalisation must be performed",
type=str2boolean,
default=False)
parser.add_argument(
"--rgb_normalisation",
help="Indicates if RGB histogram equilisation should be performed",
type=str2boolean,
default=False)
parser.add_argument(
"--whitening",
help="Indicates if the whitening of the data should be applied",
type=str2boolean,
default=False)
parser.add_argument(
"--normalise_foreground_only",
help="Indicates whether a foreground mask should be applied when"
" normalising volumes",
type=str2boolean,
default=False)
parser.add_argument("--weight_initializer",
help="Set the initializer for the weight parameters",
type=str,
default='he_normal')
parser.add_argument("--bias_initializer",
help="Set the initializer for the bias parameters",
type=str,
default='zeros')
parser.add_argument("--keep_prob",
help="Probability that each element is kept "
"if dropout is supported by the network",
type=float,
default=1.0)
yaml = require_module('yaml', mandatory=False)
if yaml:
parser.add_argument(
"--weight_initializer_args",
help="Pass arguments to the initializer for the weight parameters",
type=yaml.load,
default={})
parser.add_argument(
"--bias_initializer_args",
help="Pass arguments to the initializer for the bias parameters",
type=yaml.load,
default={})
return parser
def test_self_version(self):
require_module('importlib')
def test_installed(self):
require_module('tensorflow')
def test_no_input(self):
with self.assertRaisesRegexp(ImportError, ''):
require_module([])
with self.assertRaisesRegexp(ImportError, ''):
require_module(None)
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
from tensorflow.python.util import deprecation
deprecation._PRINT_DEPRECATION_WARNINGS = False
try:
from tensorflow.python.util import module_wrapper as deprecation
except ImportError:
from tensorflow.python.util import deprecation_wrapper as deprecation
deprecation._PER_MODULE_WARNING_LIMIT = 0
from niftynet.io.misc_io import set_logger, close_logger
set_logger()
from niftynet.utilities.util_import import require_module
require_module('blinker', descriptor='New dependency', mandatory=True)
from tensorflow.python.util import deprecation
deprecation._PRINT_DEPRECATION_WARNINGS = False
from niftynet.engine.signal import TRAIN, INFER, EVAL, EXPORT
import niftynet.utilities.util_common as util
import niftynet.utilities.user_parameters_parser as user_parameters_parser
from niftynet.engine.application_driver import ApplicationDriver
from niftynet.evaluation.evaluation_application_driver import \
EvaluationApplicationDriver
from niftynet.io.misc_io import touch_folder
from niftynet.io.misc_io import resolve_module_dir
from niftynet.io.misc_io import to_absolute_path
def imread_skimage(filename):
"""Scikit-image loader with an identity affine matrix."""
skio = require_module('skimage.io')
img = skio.imread(filename)
return image2nibabel(img)
def add_network_args(parser):
"""
keywords defined for network specification
:param parser:
:return:
"""
import niftynet.layer.binary_masking
import niftynet.layer.activation
import niftynet.utilities.histogram_standardisation as hist_std_module
parser.add_argument(
"--name",
help="Choose a net from NiftyNet/niftynet/network/ or from "
"user specified module string",
metavar='')
parser.add_argument("--activation_function",
help="Specify activation function types",
choices=list(niftynet.layer.activation.SUPPORTED_OP),
metavar='TYPE_STR',
default='relu')
parser.add_argument("--batch_size",
metavar='',
help="Set batch size of the net",
type=int,
default=2)
parser.add_argument("--decay",
help="[Training only] Set weight decay",
type=float,
default=0.0)
parser.add_argument("--reg_type",
metavar='TYPE_STR',
help="[Training only] Specify regulariser type_str",
type=str,
default='L2')
parser.add_argument(
"--volume_padding_size",
metavar='',
help="Set padding size of each volume (in all dimensions)",
type=spatialnumarray,
default=(0, 0, 0))
parser.add_argument("--window_sampling",
metavar='TYPE_STR',
help="How to sample patches from each loaded image:"
" 'uniform': fixed size uniformly distributed,"
" 'resize': resize image to the patch size.",
choices=['uniform', 'resize', 'balanced', 'weighted'],
default='uniform')
parser.add_argument("--queue_length",
help="Set size of preprocessing buffer queue",
metavar='',
type=int,
default=5)
parser.add_argument(
"--multimod_foreground_type",
choices=list(
niftynet.layer.binary_masking.SUPPORTED_MULTIMOD_MASK_TYPES),
help="Way of combining the foreground masks from different "
"modalities. 'and' is the intersection, 'or' is the union "
"and 'multi' permits each modality to use its own mask.",
default='and')
parser.add_argument(
"--histogram_ref_file",
metavar='',
type=str,
help="A reference file of histogram for intensity normalisation",
default=DEFAULT_HISTOGRAM_REF_FILE)
parser.add_argument("--norm_type",
help="Type of normalisation to perform",
type=str,
default='percentile',
choices=list(hist_std_module.SUPPORTED_CUTPOINTS))
parser.add_argument("--cutoff",
help="Cutoff values for the normalisation process",
type=float_array,
default=(0.01, 0.99))
parser.add_argument(
"--foreground_type",
choices=list(niftynet.layer.binary_masking.SUPPORTED_MASK_TYPES),
help="type_str of foreground masking strategy used",
default='otsu_plus')
parser.add_argument(
"--normalisation",
help="Indicates if the normalisation must be performed",
type=str2boolean,
default=False)
parser.add_argument(
"--whitening",
help="Indicates if the whitening of the data should be applied",
type=str2boolean,
default=False)
parser.add_argument(
"--normalise_foreground_only",
help="Indicates whether a foreground mask should be applied when"
" normalising volumes",
type=str2boolean,
default=False)
parser.add_argument("--weight_initializer",
help="Set the initializer for the weight parameters",
type=str,
default='he_normal')
parser.add_argument("--bias_initializer",
help="Set the initializer for the bias parameters",
type=str,
default='zeros')
yaml = require_module('yaml', mandatory=False)
if yaml:
parser.add_argument(
"--weight_initializer_args",
help="Pass arguments to the initializer for the weight parameters",
type=yaml.load,
default={})
parser.add_argument(
"--bias_initializer_args",
help="Pass arguments to the initializer for the bias parameters",
type=yaml.load,
default={})
return parser
def imread_skimage(filename):
"""Scikit-image loader with an identity affine matrix."""
skio = require_module('skimage.io')
img = skio.imread(filename)
return image2nibabel(img)
def imread_pillow(filename):
"""PIL (Pillow) image loader with an identity affine matrix."""
pil = require_module('PIL.Image')
img = np.asarray(pil.open(filename))
return image2nibabel(img)
def test_no_input(self):
with self.assertRaisesRegexp(ImportError, ''):
require_module([], mandatory=True)
with self.assertRaisesRegexp(ImportError, ''):
require_module(None, mandatory=True)
except AttributeError:
pass
from niftynet.utilities.versioning import get_niftynet_version_string
__version__ = get_niftynet_version_string()
import os
from niftynet.io.misc_io import set_logger
set_logger()
from niftynet.utilities.util_import import require_module
require_module('blinker', descriptor='New dependency', mandatory=True)
import niftynet.utilities.util_common as util
import niftynet.utilities.user_parameters_parser as user_parameters_parser
from niftynet.engine.application_driver import ApplicationDriver
from niftynet.evaluation.evaluation_application_driver import \
EvaluationApplicationDriver
from niftynet.io.misc_io import touch_folder
from niftynet.io.misc_io import resolve_module_dir
from niftynet.io.misc_io import to_absolute_path
def main():
system_param, input_data_param = user_parameters_parser.run()
if util.has_bad_inputs(system_param):
return -1
