def create_directories_chop(self,
base_dir,
prefix="",
as_array=False,
**kwargs):
"""
create directory structure for prepared images, and chop each image
into an image per cell.
Parameters:
-----------
base_dir : string
Path to directory in which to hold training and test datasets.
A directory will be created if it does not already exist
prefix: string
prefix for image file names
as_array: Boolean
if True will save as a numpy array. If False, then images are saved
as RGB .png files.
**kwargs: additional arguments to chop functions
"""
utils.make_dir(base_dir)
for group in self.img_dict.keys():
for key, img_list in self.img_dict[group].items():
# create directory item/key from key
dir_path = os.path.join(os.path.abspath(base_dir), group, key)
utils.make_dir(dir_path)
# create and save images in dir_path
for i, img in enumerate(img_list, 1):
rgb_img = self.convert_to_rgb(img)
# convert_to_rgb is a bit of a misnomer, actually just stacks
# an image collection to a numpy array, can work with more
# than three channels
#
# chop image into sub-img per cell
# sometimes there is an error where we don't have all the
# channel to stack into an array, not sure what is causing
# this though we can skip any errors and carry on processing
# the images rather than crash the entire session
#
# probably a much better way to handle this, but screw it
try:
sub_img_array = chop.chop_nuclei(rgb_img, **kwargs)
for j, sub_img in enumerate(sub_img_array, 1):
if as_array: # save as numpy array
img_name = "{}_img_{}_{}.npy".format(
prefix, i, j)
full_path = os.path.join(
os.path.abspath(dir_path), img_name)
np.save(file=full_path,
arr=sub_img,
allow_pickle=False)
else: # save as .png (has to be RGB)
img_name = "{}_img_{}_{}.png".format(
prefix, i, j)
full_path = os.path.join(
os.path.abspath(dir_path), img_name)
io.imsave(fname=full_path, arr=sub_img)
except ValueError:
pass
def save_chopped(arr, directory, prefix="img", ext=".png", save_as="img"):
"""
Save chopped array from chop_nuclei() to a directory. Each image will be
saved individually and consecutively numbered.
Parameters:
-----------
arr : np.ndarray
numpy array from chop_nuclei()
directory : string
directory in which to save the images.
prefix : string (default : "img")
image prefix
ext : string (default : ".png")
file extension. options are .png and .jpg if saving as an image.
Otherwise recommended extension for numpy arrays is .npy
"""
assert isinstance(arr, np.ndarray)
_check_ext_args(ext)
utils.make_dir(directory)
# loop through images in array and save with consecutive numbers
if save_as == "img":
for i, img in enumerate(arr, 1):
img_name = "{}_{}{}".format(prefix, i, ext)
full_path = os.path.join(os.path.abspath(directory), img_name)
io.imsave(fname=full_path, arr=img)
else:
for i, img in enumerate(arr, 1):
arr_name = "arr_{}.npy".format(i)
full_path = os.path.join(os.path.abspath(directory), arr_name)
np.save(full_path, img)
def create_directories_chop(self, base_dir, **kwargs):
"""
create directory structure for prepared images, and chop each image
into an image per cell.
Parameters:
-----------
base_dir : string
Path to directory in which to hold training and test datasets.
A directory will be created if it does not already exist
**kwargs: additional arguments to chop functions
"""
utils.make_dir(base_dir)
for group in self.img_dict.keys():
for key, img_list in self.img_dict[group].items():
# create directory item/key from key
dir_path = os.path.join(os.path.abspath(base_dir), group, key)
utils.make_dir(dir_path)
# create and save images in dir_path
for i, img in enumerate(img_list, 1):
rgb_img = self.convert_to_rgb(img)
# chop image into sub-img per cell
# sometimes there is an error where we don't have all the
# channel to stack into an array, not sure what is causing
# this though we can skip any errors and carry on processing
# the images rather than crash the entire session
#
# probably a much better way to handle this, but screw it
try:
sub_img_array = chop.chop_nuclei(rgb_img, **kwargs)
for j, sub_img in enumerate(sub_img_array, 1):
img_name = "img_{}_{}.npy".format(i, j)
full_path = os.path.join(os.path.abspath(dir_path),
img_name)
np.save(file=full_path, arr=sub_img)
except ValueError:
pass
def create_directories(self, base_dir):
"""
create directory structure for prepared images
Parameters:
-----------
base_dir : string
Path to directory in which to hold training and test datasets
A directory will be created if it does not already exist
"""
utils.make_dir(base_dir)
# create training and test directories
for group in self.img_dict.keys():
for key, img_list in self.img_dict[group].items():
# create directory item/key from key
dir_path = os.path.join(os.path.abspath(base_dir), group, key)
utils.make_dir(dir_path)
# create and save images in dir_path
for i, img in enumerate(img_list, 1):
# need to load images and merge
rgb_img = self.convert_to_rgb(img)
self.write_array_to_disk(img=rgb_img,
name="img_{}".format(i),
path=dir_path)
def create_directories_chop_par(self, base_dir, n_jobs=-1, size=200):
"""
create directory structure for prepared images, and chop each image
into an image per cell.
Parameters:
-----------
base_dir : string
Path to directory in which to hold training and test datasets.
A directory will be created if it does not already exist
**kwargs: additional arguments to chop functions
"""
if n_jobs < 1:
n_jobs = multiprocessing.cpu_count()
utils.make_dir(base_dir)
for group in self.img_dict.keys():
for key, img_list in self.img_dict[group].items():
# create directory item/key from key
dir_path = os.path.join(os.path.abspath(base_dir), group, key)
utils.make_dir(dir_path)
Parallel(n_jobs=n_jobs)(delayed(chopper)(img, dir_path, size)
for img in img_list)