def dataset_val(self):
"""
Returns the validation dataset. No random augmentation is performed.
:return: The validation dataset.
"""
iterator = IdListIterator(self.val_id_list_file_name,
random=False,
keys=['image_id'])
data_sources = self.data_sources(iterator, False)
if self.translate_by_random_factor:
image_size = self.image_size[:2] + [None]
else:
image_size = self.image_size
image_transformation = self.spatial_transformation(
data_sources['image_datasource'], image_size)
data_generators = self.data_generators(
data_sources['image_datasource'],
data_sources['landmarks_datasource'], image_transformation,
self.intensity_postprocessing, image_size)
dataset = GraphDataset(
data_sources=list(data_sources.values()),
data_generators=list(data_generators.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_val' if self.save_debug_images else None)
return dataset
def dataset_val(self):
"""
Returns the validation dataset. No random augmentation is performed.
:return: The validation dataset.
"""
if self.cv == 'inference':
iterator = 'iterator'
else:
iterator = self.iterator(self.test_file, False)
sources = self.datasources(iterator, False)
if self.translate_by_random_factor:
image_size = self.image_size[:2] + [None]
else:
image_size = self.image_size
reference_transformation = self.spatial_transformation(
iterator, sources, image_size)
generators = self.data_generators(iterator, sources,
reference_transformation,
self.postprocessing, False,
image_size)
if self.generate_single_vertebrae and not self.generate_labels:
del generators['labels']
return GraphDataset(
data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_val' if self.save_debug_images else None)
def dataset_val(self):
"""
Returns the validation dataset. No random augmentation is performed.
:return: The validation dataset.
"""
if self.cv == 'inference':
iterator = 'iterator'
else:
iterator = self.iterator(self.test_file, False)
sources = self.datasources(iterator, True, True, self.preprocessing, 2048)
if self.use_variable_image_size:
if self.load_spine_bbs:
image_size = ImageSizeGenerator(self.dim, [None] * 3, self.image_spacing, valid_output_sizes=[[32, 64, 96, 128], [32, 64, 96, 128], [32 + i * 32 for i in range(20)]], name='output_size', kwparents={'extent': sources['landmarks_bb_extent']})
else:
image_size = ImageSizeGenerator(self.dim, [None] * 3, self.image_spacing, valid_output_sizes=[[32, 64, 96, 128], [32, 64, 96, 128], [32 + i * 32 for i in range(20)]], name='output_size', kwparents={'image': sources['image']})
#image_size = ImageSizeGenerator(self.dim, [None] * 3, self.image_spacing, valid_output_sizes=[self.valid_output_sizes_x, self.valid_output_sizes_y, [64 + i * 32 for i in range(20)]], name='output_size', kwparents={'extent': sources['landmarks_bb_extent']})
else:
image_size = LambdaNode(lambda: self.image_size, name='output_size')
reference_transformation = self.spatial_transformation(iterator, sources, image_size)
generators = self.data_generators(iterator, sources, reference_transformation, self.postprocessing, False, image_size, False)
generators['image_id'] = LambdaNode(lambda d: np.array(d['image_id']), name='image_id', parents=[iterator])
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_val' if self.save_debug_images else None)
def dataset_val(self):
"""
Returns the validation dataset for videos. No random augmentation is performed.
:return: The validation dataset.
"""
dim = 3
full_video_frame_list_image = VideoFrameList(
self.video_frame_list_file_name,
self.num_frames - 1,
0,
border_mode='valid',
random_start=False,
random_skip_probability=0.0)
iterator = 'image_ids'
iterator_postprocessing = LambdaNode(
lambda x: full_video_frame_list_image.get_id_dict_list(
x['video_id'], x['frame_id']),
parents=[iterator])
sources = self.datasources(iterator_postprocessing)
image_key = 'image'
image_transformation = self.spatial_transformation_volumetric(
sources[image_key])
generators = self.data_generators(dim, sources, image_transformation,
None)
final_generators = self.all_generators_post_processing(
generators, False)
return GraphDataset(data_generators=list(final_generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_train'
if self.save_debug_images else None)
def dataset_val_single_frame(self):
"""
Returns the validation dataset for single frames. No random augmentation is performed.
:return: The validation dataset.
"""
sources = self.datasources_single_frame('id_dict')
image_key = 'image'
image_transformation = self.spatial_transformation(sources[image_key])
generators = self.data_generators_single_frame(2, sources, image_transformation, None)
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator='id_dict',
debug_image_folder=self.debug_image_folder)
def dataset_val_single_frame(self):
"""
Returns the validation dataset for single frames. No random augmentation is performed.
:return: The validation dataset.
"""
sources = self.datasources_single_frame('id_dict')
image_key = 'merged' if self.pad_image or self.crop_image_size is not None else 'image'
image_transformation = self.spatial_transformation(sources[image_key])
generators = self.data_generators_single_frame(2, sources, image_transformation, None)
final_generators = self.all_generators_post_processing(generators, False)
return GraphDataset(data_generators=list(final_generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator='id_dict',
debug_image_folder='debug_val' if self.save_debug_images else None)
def dataset_single_frame(self):
"""
Returns the training dataset for single frames. Random augmentation is performed.
:return: The training dataset.
"""
iterator = 'iterator'
sources = self.datasources_single_frame(iterator)
image_key = 'merged'
image_transformation = self.spatial_transformation(sources[image_key])
generators = self.data_generators_single_frame(2, sources,
image_transformation)
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder=None)
def dataset_val_all_frames(self):
"""
Returns the validation dataset for videos. No random augmentation is performed.
:return: The validation dataset.
"""
dim = 3
iterator = 'image_ids'
sources = self.datasources(iterator)
image_key = 'merged' if self.pad_image or self.crop_image_size is not None else 'image'
image_transformation = self.spatial_transformation_volumetric(sources[image_key])
generators = self.data_generators(dim, sources, image_transformation, None)
final_generators = self.all_generators_post_processing(generators, False)
return GraphDataset(data_generators=list(final_generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_train' if self.save_debug_images else None)
def dataset_train_single_frame(self):
"""
Returns the training dataset for single frames. Random augmentation is performed.
:return: The training dataset.
"""
iterator = IdListIterator(self.train_id_list_file_name, random=True, keys=['video_id', 'frame_id'])
sources = self.datasources_single_frame(iterator)
image_key = 'merged' if self.pad_image or self.crop_image_size is not None else 'image'
image_transformation = self.spatial_transformation_augmented(sources[image_key])
generators = self.data_generators_single_frame(2, sources, image_transformation, self.postprocessing_random)
final_generators = self.all_generators_post_processing(generators, False)
return GraphDataset(data_generators=list(final_generators.values()),
data_sources=list(sources.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_train' if self.save_debug_images else None)
def dataset_val(self):
"""
Returns the validation dataset. No random augmentation is performed.
:return: The validation dataset.
"""
iterator = IdListIterator(self.val_id_list_file_name,
random=False,
keys=['image_id'])
data_sources = self.data_sources(False, iterator)
image_transformation = self.spatial_transformation(data_sources)
data_generators = self.data_generators(data_sources,
image_transformation,
self.intensity_postprocessing)
return GraphDataset(
data_generators=list(data_generators.values()),
data_sources=list(data_sources.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_val' if self.save_debug_images else None)
def dataset_train(self):
"""
Returns the training dataset. Random augmentation is performed.
:return: The training dataset.
"""
iterator = IdListIterator(self.train_file,
random=True,
keys=['image_id'],
name='iterator')
sources = self.datasources(iterator, True)
reference_transformation = self.spatial_transformation_augmented(
sources)
generators = self.data_generators(sources, reference_transformation,
self.postprocessing_random)
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_train'
if self.save_debug_images else None)
def dataset_train(self):
"""
Returns the training dataset. Random augmentation is performed.
:return: The training dataset.
"""
iterator = IdListIterator(self.train_id_list_file_name,
random=True,
keys=['image_id'])
data_sources = self.data_sources(iterator, True)
image_transformation = self.spatial_transformation_augmented(
data_sources['image_datasource'])
data_generators = self.data_generators(
data_sources['image_datasource'],
data_sources['landmarks_datasource'], image_transformation,
self.intensity_postprocessing_augmented, self.image_size)
dataset = GraphDataset(data_sources=list(data_sources.values()),
data_generators=list(data_generators.values()),
transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_train'
if self.save_debug_images else None)
return dataset
def dataset_train(self):
"""
Returns the training dataset for videos. Random augmentation is performed.
:return: The training dataset.
"""
dim = 3
full_video_frame_list_image = VideoFrameList(self.video_frame_list_file_name, self.num_frames - 1, 0,
border_mode='valid', random_start=True, random_skip_probability=self.random_skip_probability)
iterator = IdListIterator(self.train_id_list_file_name, random=True, keys=['video_id', 'frame_id'],
postprocessing=lambda x: full_video_frame_list_image.get_id_dict_list(x['video_id'], x['frame_id']))
sources = self.datasources(iterator)
image_key = 'merged' if self.pad_image or self.crop_image_size is not None else 'image'
image_transformation = self.spatial_transformation_volumetric_augmented(sources[image_key])
generators = self.data_generators(dim, sources, image_transformation, self.postprocessing_random)
final_generators = self.all_generators_post_processing(generators, False)
return GraphDataset(data_generators=list(final_generators.values()),
# data_sources=list(sources.values()),
# transformations=[image_transformation],
iterator=iterator,
debug_image_folder='debug_train' if self.save_debug_images else None)
def dataset_train(self):
"""
Returns the training dataset. Random augmentation is performed.
:return: The training dataset.
"""
iterator = self.iterator(self.train_file, True)
sources = self.datasources(iterator, False, False, self.preprocessing_random, 8192)
if self.use_variable_image_size:
image_size = ImageSizeGenerator(self.dim, [None] * 3, self.image_spacing, valid_output_sizes=[self.valid_output_sizes_x, self.valid_output_sizes_y, self.valid_output_sizes_z], name='output_size', kwparents={'extent': sources['landmarks_bb_extent']})
if self.crop_randomly_smaller:
image_size = LambdaNode(self.crop_randomly_smaller_image_size, name='output_size', parents=[image_size])
else:
image_size = LambdaNode(lambda: self.image_size, name='output_size')
reference_transformation = self.spatial_transformation_augmented(iterator, sources, image_size)
generators = self.data_generators(iterator, sources, reference_transformation, self.postprocessing_random, True, image_size, self.crop_image_top_bottom)
generators['image_id'] = LambdaNode(lambda d: np.array(d['image_id']), name='image_id', parents=[iterator])
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_train' if self.save_debug_images else None)
def dataset_train(self):
"""
Returns the training dataset. Random augmentation is performed.
:return: The training dataset.
"""
iterator = self.iterator(self.train_file, True)
sources = self.datasources(iterator, True)
image_size = self.image_size
reference_transformation = self.spatial_transformation_augmented(
iterator, sources, image_size)
generators = self.data_generators(iterator, sources,
reference_transformation,
self.postprocessing_random, True,
image_size)
if self.generate_single_vertebrae and not self.generate_labels:
del generators['labels']
return GraphDataset(data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_train'
if self.save_debug_images else None)
def dataset_val(self):
"""
Returns the validation dataset. No random augmentation is performed.
:return: The validation dataset.
"""
iterator = IdListIterator(self.test_file,
random=False,
keys=['image_id'],
name='iterator')
sources = self.datasources(iterator, False)
reference_transformation = self.spatial_transformation(sources)
generators = self.data_generators(sources, reference_transformation,
self.postprocessing)
if self.cv == 0:
del sources['landmarks']
del generators['landmarks']
return GraphDataset(
data_generators=list(generators.values()),
data_sources=list(sources.values()),
transformations=[reference_transformation],
iterator=iterator,
debug_image_folder='debug_val' if self.save_debug_images else None)