def generate_correct(image, prob, segmentation):
hist = Util.get_histogram(segmentation.astype(np.uint64))
labels = range(1, len(hist)) # do not include zeros
np.random.shuffle(labels)
for l in labels:
binary_mask = Util.threshold(segmentation, l)
borders = mh.labeled.borders(binary_mask, Bc=mh.disk(2))
labeled_borders = skimage.measure.label(borders)
labeled_borders[borders == 0] = 0
relabeled_borders, no_relabeled_borders = mh.labeled.relabel(
labeled_borders.astype(np.uint16))
for border in range(1, no_relabeled_borders + 1):
isolated_border = Util.threshold(relabeled_borders, border)
patches = Patch.analyze_border(image, prob, binary_mask,
binary_mask.invert(),
isolated_border)
for s in patches:
yield s
yield Patch.fliplr(s)
yield Patch.flipud(s)
yield Patch.rotate(s, 90)
yield Patch.rotate(s, 180)
yield Patch.rotate(s, 270)
def generate_split_error(image, prob, segmentation, n=3):
hist = Util.get_histogram(segmentation.astype(np.uint64))
labels = range(1, len(hist)) # do not include zeros
np.random.shuffle(labels)
for l in labels:
binary_mask = Util.threshold(segmentation, l)
labeled_parts = skimage.measure.label(binary_mask)
labeled_parts += 1 # avoid the 0
labeled_parts[binary_mask == 0] = 0
labeled_parts, no_labeled_parts = mh.labeled.relabel(labeled_parts)
for i in range(n):
for part in range(1, no_labeled_parts + 1):
binary_part = Util.threshold(labeled_parts, part)
split_binary_mask, split_isolated_border = Patch.split_label(
image, binary_part)
if split_binary_mask.max() == 0:
# the strange err (label too small or sth)
print 'Caught empty..'
continue
patches = Patch.analyze_border(image, prob,
split_binary_mask,
binary_mask.invert(),
split_isolated_border)
for s in patches:
yield s
yield Patch.fliplr(s)
yield Patch.flipud(s)
yield Patch.rotate(s, 90)
yield Patch.rotate(s, 180)
yield Patch.rotate(s, 270)