def test_match_non_unitary_at_01(self):
matcher = MatchEngineIoU(0.1, 'non-unitary')
assert np.all(
matcher.match(detections, gt) == np.array(
[[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 1, 1], [0, 0, 0, 0, 1, 0],
[0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0], [0, 0, 1, 0, 0, 0]]))
def test_match_xview_at_005(self):
matcher = MatchEngineIoU(0.05, 'xview')
assert np.all(
matcher.match(detections, gt, label_mean_area=gt_mean_area) == np.
array([[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 1], [0, 0, 0, 0, 1, 0],
[0, 0, 0, 1, 0, 0], [1, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0]]))
def test_match_coco_at_005(self):
matcher = MatchEngineIoU(0.05, 'coco')
assert np.all(
matcher.match(detections, gt) == np.array(
[[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 1], [0, 0, 0, 0, 1, 0],
[0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0], [0, 1, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0], [0, 0, 0, 0, 0, 0]]))
def test_match_xview_at_01(self):
matcher = MatchEngineIoU(0.1, 'xview')
assert np.all(
matcher.match(detections, gt) == np.array(
[[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 1], [0, 0, 0, 0, 1, 0],
[0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0], [0, 0, 1, 0, 0, 0]]))
def test_match_coco_at_01(self):
matcher = MatchEngineIoU(0.1, 'coco')
assert np.all(
matcher.match(detections, gt, label_mean_area=gt_mean_area) == np.
array([[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 1], [0, 0, 0, 0, 1, 0],
[0, 0, 0, 1, 0, 0], [1, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0]]))
def test_rtree_iou_matrix(self):
matcher = MatchEngineIoU(0.1, 'coco')
ref_iou = naive_compute_iou_matrix(sort_detection_by_confidence(detections), gt)
iou = matcher.compute_similarity_matrix(detections, gt)
print(iou)
print(ref_iou)
assert np.all(iou == ref_iou)
def test_rtree_iou_matrix(self):
matcher = MatchEngineIoU(0.1, 'coco')
ref_iou = naive_compute_iou_matrix(sort_detection_by_confidence(detections), gt) * \
(gt_mean_area / area(gt[:, 0]))
iou = matcher.compute_similarity_matrix(detections, gt, label_mean_area=gt_mean_area)
print(iou)
print(ref_iou)
assert np.all(iou[np.logical_not(np.isinf(iou))] == ref_iou[np.logical_not(np.isinf(iou))])
def test_threshold_property(self):
class MatchEngine(MatchEngineBase):
def compute_similarity_matrix(self, detections, ground_truths, label_mean_area=None):
pass
def trim_similarity_matrix(self, similarity_matrix, detections, ground_truths, label_mean_area=None):
pass
map_with_threshold = MeanAveragePrecisionMetric(0.5, 'coco', match_engine=MatchEngineIoU(0.5, 'coco'))
map_without_threshold = MeanAveragePrecisionMetric(0.5, 'coco', match_engine=MatchEngine('coco'))
assert map_with_threshold.threshold == 0.5
assert map_without_threshold.threshold is None
def test_match_xview_at_01(self):
matcher_poly = MatchEngineIoU(0.1, 'xview')
matcher_bbox = MatchEngineIoU(0.1, 'xview')
assert \
np.all(matcher_bbox.match(detections_bbox,
gt_bbox,
label_mean_area=gt_mean_area) == matcher_poly.match(detections_poly,
gt_poly,
label_mean_area=gt_mean_area))
def test_match_coco_at_005(self):
matcher_poly = MatchEngineIoU(0.05, 'coco')
matcher_bbox = MatchEngineIoU(0.05, 'coco')
assert \
np.all(matcher_bbox.match(detections_bbox,
gt_bbox,
label_mean_area=gt_mean_area) == matcher_poly.match(detections_poly,
gt_poly,
label_mean_area=gt_mean_area))
def test_rtree_iou_matrix(self):
matcher_poly = MatchEngineIoU(0.1, 'coco')
matcher_bbox = MatchEngineIoU(0.1, 'coco')
iou_poly = matcher_poly.compute_similarity_matrix(
detections_corr_poly, gt_corr_poly, label_mean_area=gt_mean_area)
iou_bbox = matcher_bbox.compute_similarity_matrix(
detections_bbox, gt_bbox, label_mean_area=gt_mean_area)
print(iou_poly)
print(iou_bbox)
assert np.all(iou_poly == iou_bbox)
def test_match_xview_at_01(self):
matcher_poly = MatchEngineIoU(0.1, 'xview')
matcher_bbox = MatchEngineIoU(0.1, 'xview')
assert np.all(
matcher_bbox.match(detections_bbox, gt_bbox) == matcher_poly.match(
detections_poly, gt_poly))
def test_match_algorithm_warning(self):
with pytest.warns(RuntimeWarning, match='Discrepancy between user provided match_algorithm'):
_ = MeanAveragePrecisionMetric(0.5, 'xview', match_engine=MatchEngineIoU(0.5, 'coco'))
def _make_scale_match(self, threshold, method):
matcher = MatchEngineIoU(threshold, method)
matcher.match(self.data_detect_conv,
self.data_detect_conv[:, (True, False, True)])
def test_match_non_unitary_at_01(self):
matcher_poly = MatchEngineIoU(0.1, 'non-unitary')
matcher_bbox = MatchEngineIoU(0.1, 'non-unitary')
assert np.all(
matcher_bbox.match(detections_bbox, gt_bbox) == matcher_poly.match(
detections_poly, gt_poly))
def test_match_coco_at_005(self):
matcher_poly = MatchEngineIoU(0.05, 'coco')
matcher_bbox = MatchEngineIoU(0.05, 'coco')
assert np.all(
matcher_bbox.match(detections_bbox, gt_bbox) == matcher_poly.match(
detections_poly, gt_poly))
def test_threshold_warning(self):
import warnings
warnings.filterwarnings("always", category=RuntimeWarning)
class MatchEngine(MatchEngineBase):
def compute_similarity_matrix(self, detections, ground_truths, label_mean_area=None):
pass
def trim_similarity_matrix(self, similarity_matrix, detections, ground_truths, label_mean_area=None):
pass
with pytest.warns(RuntimeWarning, match='Discrepancy between user provided threshold'):
map_with_threshold = MeanAveragePrecisionMetric(0.1, 'coco', match_engine=MatchEngineIoU(0.5, 'coco'))
with pytest.warns(RuntimeWarning, match='Discrepancy between user provided threshold'):
map_without_threshold = MeanAveragePrecisionMetric(0.1, 'coco', match_engine=MatchEngine('coco'))