def test_get_image_sizes(self):
temp_dir = self.get_temp_dir()
_write_image(os.path.join(temp_dir, 'image1.png'), 300, 300)
_write_image(os.path.join(temp_dir, 'image2.png'), 300, 300)
# test with single channel name
size = io_utils.get_image_sizes(temp_dir, ['image1'])
self.assertEqual(size, (300, 300))
# test with multiple channel names
sizes = io_utils.get_image_sizes(temp_dir, ['image1', 'image2'])
self.assertEqual(sizes, (300, 300))
def make_training_data_3d(direc_name,
file_name_save,
channel_names,
training_direcs=None,
annotation_name='corrected',
raw_image_direc='raw',
annotation_direc='annotated',
reshape_size=None,
num_frames=50,
montage_mode=True):
"""
Read all images in training directories and save as npz file
3D image sets are "stacks" of images. For annotation purposes, these images
have been sliced into "montages", where a section of each stack has been
sliced for efficient annotated by humans. The raw_image_direc should be a
specific montage (e.g. montage_0_0) and the annotation is the corresponding
annotated montage.
# Arguments
direc_name: directory containing folders of training data
file_name_save: full filepath for npz file where the data will be saved
training_direcs: directories of images located inside direc_name
If None, all directories in direc_name are used.
channel_names: Loads all raw images with a channel_name in the filename
raw_image_direc: directory name inside each training dir with raw images
annotation_direc: directory name inside each training dir with masks
reshape_size: If provided, will reshape the images to the given size.
num_frames: number of frames to load from each training directory
montage_mode: load masks from "montaged" subdirs inside annotation_direc
"""
# Load one file to get image sizes
rand_train_dir = os.path.join(direc_name, random.choice(training_direcs), raw_image_direc)
if montage_mode:
rand_train_dir = os.path.join(rand_train_dir, random.choice(os.listdir(rand_train_dir)))
image_size = get_image_sizes(rand_train_dir, channel_names)
X = load_training_images_3d(direc_name, training_direcs,
raw_image_direc=raw_image_direc,
channel_names=channel_names,
image_size=image_size,
num_frames=num_frames,
montage_mode=montage_mode)
y = load_annotated_images_3d(direc_name, training_direcs,
annotation_direc=annotation_direc,
annotation_name=annotation_name,
image_size=image_size,
num_frames=num_frames,
montage_mode=montage_mode)
# Reshape X and y
if reshape_size is not None:
X, y = reshape_movie(X, y, reshape_size=reshape_size)
np.savez(file_name_save, X=X, y=y)
return None
def make_training_data_2d(direc_name,
file_name_save,
channel_names,
raw_image_direc='raw',
annotation_direc='annotated',
annotation_name='feature',
training_direcs=None,
reshape_size=None):
"""Read all images in training directories and save as npz file.
Args:
direc_name (str): directory containing folders of training data
file_name_save (str): full filepath for npz file
where the data will be saved
training_direcs (str[]): directories of images located inside
direc_name. If None, all directories in direc_name are used.
raw_image_direc (str): directory name inside each training dir
with raw images
annotation_direc (str): directory name inside each training dir
with masks
channel_names (str[]): Loads all raw images with
a channel_name in the filename
annotation_name (str): Loads all masks with
annotation_name in the filename
reshape_size (int): If provided, reshapes the images to the given size
"""
# Load one file to get image sizes (assumes all images same size)
image_path = os.path.join(direc_name, random.choice(training_direcs),
raw_image_direc)
image_size = get_image_sizes(image_path, channel_names)
X = load_training_images_2d(direc_name,
training_direcs,
raw_image_direc=raw_image_direc,
channel_names=channel_names,
image_size=image_size)
y = load_annotated_images_2d(direc_name,
training_direcs,
annotation_direc=annotation_direc,
annotation_name=annotation_name,
image_size=image_size)
if reshape_size is not None:
X, y = reshape_matrix(X, y, reshape_size=reshape_size)
# Save training data in npz format
np.savez(file_name_save, X=X, y=y)
def make_training_data_3d(direc_name,
file_name_save,
channel_names,
training_direcs=None,
annotation_name='corrected',
raw_image_direc='raw',
annotation_direc='annotated',
reshape_size=None,
num_frames=None,
montage_mode=True):
"""Read all images in training directories and save as npz file.
3D image sets are "stacks" of images. For annotation purposes, these images
have been sliced into "montages", where a section of each stack has been
sliced for efficient annotated by humans. The raw_image_direc should be a
specific montage (e.g. montage_0_0) and the annotation is the corresponding
annotated montage.
Args:
direc_name (str): directory containing folders of training data
file_name_save (str): full filepath for npz file where the
data will be saved
training_direcs (str[]): directories of images located inside
direc_name. If None, all directories in direc_name are used.
annotation_name (str): Loads all masks with
annotation_name in the filename
channel_names (str[]): Loads all raw images with
a channel_name in the filename
raw_image_direc (str): directory name inside each
training dir with raw images
annotation_direc (str): directory name inside each
training dir with masks
reshape_size (int): If provided, reshapes the images to the given size.
num_frames (int): number of frames to load from each training directory
montage_mode (bool): load masks from "montaged" subdirs
inside annotation_direc
"""
# Load one file to get image sizes
rand_train_dir = os.path.join(direc_name, random.choice(training_direcs),
raw_image_direc)
if montage_mode:
rand_train_dir = os.path.join(
rand_train_dir, random.choice(os.listdir(rand_train_dir)))
image_size = get_image_sizes(rand_train_dir, channel_names)
if num_frames is None:
raw = [
os.path.join(direc_name, t, raw_image_direc)
for t in training_direcs
]
ann = [
os.path.join(direc_name, t, annotation_direc)
for t in training_direcs
]
if montage_mode:
raw = [os.path.join(r, d) for r in raw for d in os.listdir(r)]
ann = [os.path.join(a, d) for a in ann for d in os.listdir(a)]
# use all images if not set
# will select the first N images where N is the smallest
# number of images in each subdir of all training_direcs
num_frames_raw = min([count_image_files(f) for f in raw])
num_frames_ann = min([count_image_files(f) for f in ann])
num_frames = min(num_frames_raw, num_frames_ann)
X = load_training_images_3d(direc_name,
training_direcs,
raw_image_direc=raw_image_direc,
channel_names=channel_names,
image_size=image_size,
num_frames=num_frames,
montage_mode=montage_mode)
y = load_annotated_images_3d(direc_name,
training_direcs,
annotation_direc=annotation_direc,
annotation_name=annotation_name,
image_size=image_size,
num_frames=num_frames,
montage_mode=montage_mode)
# Reshape X and y
if reshape_size is not None:
X, y = reshape_movie(X, y, reshape_size=reshape_size)
np.savez(file_name_save, X=X, y=y)
