def __init__(self,
X,
Y,
n_rays,
grid,
batch_size,
patch_size,
use_gpu=False,
sample_ind_cache=True,
maxfilter_patch_size=None,
augmenter=None,
foreground_prob=0):
X = [x.astype(np.float32, copy=False) for x in X]
# Y = [y.astype(np.uint16, copy=False) for y in Y]
# sanity checks
assert len(X) == len(Y) and len(X) > 0
nD = len(patch_size)
assert nD in (2, 3)
x_ndim = X[0].ndim
assert x_ndim in (nD, nD + 1)
assert all(
y.ndim == nD and x.ndim == x_ndim and x.shape[:nD] == y.shape
for x, y in zip(X, Y))
if x_ndim == nD:
self.n_channel = None
else:
self.n_channel = X[0].shape[-1]
assert all(x.shape[-1] == self.n_channel for x in X)
assert 0 <= foreground_prob <= 1
self.X, self.Y = X, Y
self.batch_size = batch_size
self.n_rays = n_rays
self.patch_size = patch_size
self.ss_grid = (slice(None), ) + tuple(slice(0, None, g) for g in grid)
self.perm = np.random.permutation(len(self.X))
self.use_gpu = bool(use_gpu)
if augmenter is None:
augmenter = lambda *args: args
callable(augmenter) or _raise(
ValueError("augmenter must be None or callable"))
self.augmenter = augmenter
self.foreground_prob = foreground_prob
if self.use_gpu:
from gputools import max_filter
self.max_filter = lambda y, patch_size: max_filter(
y.astype(np.float32), patch_size)
else:
from scipy.ndimage.filters import maximum_filter
self.max_filter = lambda y, patch_size: maximum_filter(
y, patch_size, mode='constant')
self.maxfilter_patch_size = maxfilter_patch_size if maxfilter_patch_size is not None else self.patch_size
self.sample_ind_cache = sample_ind_cache
self._ind_cache_fg = {}
self._ind_cache_all = {}
def __init__(self, X, Y, n_params, grid, batch_size, patch_size, length, use_gpu=False, sample_ind_cache=True, maxfilter_patch_size=None, augmenter=None, foreground_prob=0):
super().__init__(data_size=len(X), batch_size=batch_size, length=length, shuffle=True)
if isinstance(X, (np.ndarray, tuple, list)):
X = [x.astype(np.float32, copy=False) for x in X]
# Y = [y.astype(np.uint16, copy=False) for y in Y]
# sanity checks
assert len(X)==len(Y) and len(X)>0
nD = len(patch_size)
assert nD in (2,3)
x_ndim = X[0].ndim
assert x_ndim in (nD,nD+1)
if isinstance(X, (np.ndarray, tuple, list)) and \
isinstance(Y, (np.ndarray, tuple, list)):
all(y.ndim==nD and x.ndim==x_ndim and x.shape[:nD]==y.shape for x,y in zip(X,Y)) or _raise("images and masks should have corresponding shapes/dimensions")
#REFACTORED
# all(x.shape[:nD]>=patch_size for x in X) or _raise("Some images are too small for given patch_size {patch_size}".format(patch_size=patch_size))
if x_ndim == nD:
self.n_channel = None
else:
self.n_channel = X[0].shape[-1]
assert all(x.shape[-1]==self.n_channel for x in X)
assert 0 <= foreground_prob <= 1
self.X, self.Y = X, Y
# self.batch_size = batch_size
self.n_params = n_params
self.patch_size = patch_size
self.ss_grid = (slice(None),) + tuple(slice(0, None, g) for g in grid)
self.use_gpu = bool(use_gpu)
if augmenter is None:
augmenter = lambda *args: args
callable(augmenter) or _raise(ValueError("augmenter must be None or callable"))
self.augmenter = augmenter
self.foreground_prob = foreground_prob
if self.use_gpu:
from gputools import max_filter
self.max_filter = lambda y, patch_size: max_filter(y.astype(np.float32), patch_size)
else:
from scipy.ndimage.filters import maximum_filter
self.max_filter = lambda y, patch_size: maximum_filter(y, patch_size, mode='constant')
self.maxfilter_patch_size = maxfilter_patch_size if maxfilter_patch_size is not None else self.patch_size
self.sample_ind_cache = sample_ind_cache
self._ind_cache_fg = {}
self._ind_cache_all = {}
def __init__(self,
X,
Y,
n_rays,
grid,
batch_size,
patch_size,
use_gpu=False,
maxfilter_cache=True,
maxfilter_patch_size=None,
augmenter=None):
X = [x.astype(np.float32, copy=False) for x in X]
# Y = [y.astype(np.uint16, copy=False) for y in Y]
self.X, self.Y = X, Y
self.batch_size = batch_size
self.n_rays = n_rays
self.patch_size = patch_size
self.ss_grid = (slice(None), ) + tuple(slice(0, None, g) for g in grid)
self.perm = np.random.permutation(len(self.X))
self.use_gpu = bool(use_gpu)
if augmenter is None:
augmenter = lambda *args: args
callable(augmenter) or _raise(
ValueError("augmenter must be None or callable"))
self.augmenter = augmenter
if self.use_gpu:
from gputools import max_filter
self.max_filter = lambda y, patch_size: max_filter(
y.astype(np.float32), patch_size)
else:
from scipy.ndimage.filters import maximum_filter
self.max_filter = lambda y, patch_size: maximum_filter(
y, patch_size, mode='constant')
self.maxfilter_patch_size = maxfilter_patch_size if maxfilter_patch_size is not None else self.patch_size
if maxfilter_cache:
self.R = [
self.no_background_patches((x, y))
for x, y in zip(self.X, self.Y)
]
else:
self.R = None