def result_per_category(self):
"""See base class."""
sq = base_metric.realdiv_maybe_zero(self.iou_per_class,
self.tp_per_class)
rq = base_metric.realdiv_maybe_zero(
self.tp_per_class, self.tp_per_class + 0.5 * self.fn_per_class +
0.5 * self.fp_per_class)
return np.multiply(sq, rq)
def detailed_results(self, is_thing=None):
"""See base class."""
valid_categories = self._valid_categories()
# If known, break down which categories are valid _and_ things/stuff.
category_sets = collections.OrderedDict()
category_sets['All'] = valid_categories
if is_thing is not None:
category_sets['Things'] = np.logical_and(valid_categories,
is_thing)
category_sets['Stuff'] = np.logical_and(valid_categories,
np.logical_not(is_thing))
# Compute individual per-class metrics that constitute factors of PQ.
sq = base_metric.realdiv_maybe_zero(self.iou_per_class,
self.tp_per_class)
rq = base_metric.realdiv_maybe_zero(
self.tp_per_class, self.tp_per_class + 0.5 * self.fn_per_class +
0.5 * self.fp_per_class)
pq = np.multiply(sq, rq)
# Assemble detailed results dictionary.
results = {}
for category_set_name, in_category_set in six.iteritems(category_sets):
if np.any(in_category_set):
results[category_set_name] = {
'pq': np.mean(pq[in_category_set]),
'sq': np.mean(sq[in_category_set]),
'rq': np.mean(rq[in_category_set]),
# The number of categories in this subset.
'n': np.sum(in_category_set.astype(np.int32)),
}
else:
results[category_set_name] = {
'pq': 0,
'sq': 0,
'rq': 0,
'n': 0
}
return results
def result_per_category(self):
"""See base class."""
return base_metric.realdiv_maybe_zero(self.weighted_iou_per_class,
self.gt_area_per_class)
def result_per_category(self):
"""See base class."""
return base_metric.realdiv_maybe_zero(self.weighted_iou_per_class,
self.gt_area_per_class)
