def test_multiple_detections_with_ignore_difficult_and_not_allow_multiple_matches_per_ignored(
self):
gt = make_representation("0 0 0 5 5", is_ground_truth=True)
pred = make_representation("1 0 0 0 5 5; 0.9 0 0 0 5 5")
gt[0].metadata['difficult_boxes'] = [0]
overlap_evaluator = IOU({})
tp, fp, _, n = bbox_match(gt,
pred,
0,
overlap_evaluator,
ignore_difficult=True,
allow_multiple_matches_per_ignored=False)[:4]
assert n == 0
assert tp[0] == 0
assert tp[1] == 0
assert fp[0] == 0
assert fp[1] == 1
def test_two_objects(self):
gt = make_representation(["0 0 0 5 5; 0 10 10 20 20"],
is_ground_truth=True)
pred = make_representation(["0 0 0 5 5; 0 10 10 20 20"], score=1)
assert 1 == _test_metric_wrapper(Recall,
single_class_dataset())(gt, pred)[0]
def test_one_object(self):
gt = make_representation(["0 0 0 5 5"], is_ground_truth=True)
pred = make_representation(["0 0 0 5 5"], score=1)
metric = _test_metric_wrapper(Recall, single_class_dataset())
assert 1 == metric(gt, pred)[0]
assert metric.meta.get('names') == ['dog']
def test_false_negative(self):
gt = make_representation(["0 10 10 20 20; 0 0 0 5 5"], is_ground_truth=True)
pred = make_representation(["0 0 0 5 5"], score=1)
metric = _test_metric_wrapper(Recall, single_class_dataset())
assert 0.5 == metric(gt, pred)[0]
assert metric.meta.get('names') == ['dog']