def __call__(self, sample):
"""Affinity label processing."""
seg = sample[self.source]
msk = get_mask(sample, self.source)
affs = list()
msks = list()
# Affinitize.
for dst in self.dst:
affs.append(tf.affinitize1(seg, dst=dst))
msks.append(tf.affinitize1_mask(msk, dst=dst))
aff = np.concatenate(affs, axis=0)
msk = np.concatenate(msks, axis=0)
# Rebalancing.
if self.rebalance:
for c in xrange(aff.shape[0]):
msk[c, ...] *= tf.rebalance_binary_class(aff[c, ...],
msk=msk[c, ...])
# Update sample.
sample[self.target] = aff
sample[self.target + '_mask'] = msk
# Crop.
if self.crop is not None:
for k, v in sample.iteritems():
sample[k] = tf.crop(v, offset=self.crop)
return sample
def __call__(self, sample):
"""Synapse label processing."""
syn = sample[self.source]
# Binarize.
lbl = tf.binarize(syn)
msk = get_mask(sample, self.source)
# Rebalancing.
if self.rebalance:
msk = tf.rebalance_binary_class(lbl, msk, base_w=self.base_w)
# Update sample.
sample[self.target] = lbl
sample[self.target + '_mask'] = msk
return sample
def __call__(self, sample):
"""Semantic label processing."""
sem = sample[self.source]
# Semantic class expansion.
lbl, msk = tf.multiclass_expansion(sem, ids=self.ids)
# Combine with a given mask.
msk *= get_mask(sample, self.source)
# Rebalancing.
if self.rebalance:
for i, _ in enumerate(self.ids):
msk[i,
...] = tf.rebalance_binary_class(lbl[i, ...], msk[i, ...])
# Update sample.
sample[self.target] = lbl
sample[self.target + '_mask'] = msk
return sample