def setUp(self):
self.class_map = ClassMap([ClassItem(1, 'car'), ClassItem(2, 'house')])
self.npboxes = np.array([
[0., 0., 2., 2.],
[2., 2., 4., 4.],
])
self.class_ids = np.array([1, 2])
self.scores = np.array([0.9, 0.9])
self.labels = ObjectDetectionLabels(self.npboxes,
self.class_ids,
scores=self.scores)
def test_filter_by_aoi(self):
aois = [Box.make_square(0, 0, 2).to_shapely()]
filt_labels = self.labels.filter_by_aoi(aois)
npboxes = np.array([[0., 0., 2., 2.]])
class_ids = np.array([1])
scores = np.array([0.9])
exp_labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
self.assertEqual(filt_labels, exp_labels)
aois = [Box.make_square(4, 4, 2).to_shapely()]
filt_labels = self.labels.filter_by_aoi(aois)
exp_labels = ObjectDetectionLabels.make_empty()
self.assertEqual(filt_labels, exp_labels)
def test_concatenate(self):
npboxes = np.array([[4., 4., 5., 5.]])
class_ids = np.array([2])
scores = np.array([0.3])
labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
new_labels = ObjectDetectionLabels.concatenate(self.labels, labels)
npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.],
[4., 4., 5., 5.]])
class_ids = np.array([1, 2, 2])
scores = np.array([0.9, 0.9, 0.3])
expected_labels = ObjectDetectionLabels(
npboxes, class_ids, scores=scores)
new_labels.assert_equal(expected_labels)
def test_constructor(self):
labels = ObjectDetectionLabels(self.npboxes,
self.class_ids,
scores=self.scores)
expected_labels = ObjectDetectionLabels(self.npboxes, self.class_ids,
self.scores)
labels.assert_equal(expected_labels)
labels = ObjectDetectionLabels(self.npboxes, self.class_ids)
scores = np.ones(self.class_ids.shape)
expected_labels = ObjectDetectionLabels(self.npboxes,
self.class_ids,
scores=scores)
labels.assert_equal(expected_labels)
def test_from_boxlist(self):
from rastervision.data.label.tfod_utils.np_box_list import BoxList
boxlist = BoxList(self.npboxes)
boxlist.add_field('classes', self.class_ids)
boxlist.add_field('scores', self.scores)
labels = ObjectDetectionLabels.from_boxlist(boxlist)
labels.assert_equal(self.labels)
def test_normalized_to_local(self):
norm_npboxes = np.array([[0., 0., 0.2, 0.02], [0.2, 0.02, 0.4, 0.04]])
window = Box(0, 0, 10, 100)
local_npboxes = ObjectDetectionLabels.normalized_to_local(
norm_npboxes, window)
expected_local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
np.testing.assert_array_equal(local_npboxes, expected_local_npboxes)
def test_global_to_local(self):
global_npboxes = np.array([[10., 10., 12., 12.], [12., 12., 14., 14.]])
window = Box.make_square(10, 10, 10)
local_npboxes = ObjectDetectionLabels.global_to_local(
global_npboxes, window)
expected_local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
np.testing.assert_array_equal(local_npboxes, expected_local_npboxes)
def test_prune_duplicates(self):
# This first box has a score below score_thresh so it should get
# pruned. The third box overlaps with the second, but has higher score,
# so the second one should get pruned. The fourth box overlaps with
# the second less than merge_thresh, so it should not get pruned.
npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.],
[2.1, 2.1, 4.1, 4.1], [3.5, 3.5, 5.5, 5.5]])
class_ids = np.array([1, 2, 1, 2])
scores = np.array([0.2, 0.9, 0.9, 1.0])
labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
score_thresh = 0.5
merge_thresh = 0.5
pruned_labels = ObjectDetectionLabels.prune_duplicates(
labels, score_thresh, merge_thresh)
self.assertEqual(len(pruned_labels), 2)
expected_npboxes = np.array([[2.1, 2.1, 4.1, 4.1],
[3.5, 3.5, 5.5, 5.5]])
expected_class_ids = np.array([1, 2])
expected_scores = np.array([0.9, 1.0])
# prune_duplicates does not maintain ordering of boxes, so find match
# between pruned boxes and expected_npboxes.
pruned_npboxes = pruned_labels.get_npboxes()
pruned_inds = [None, None]
for box_ind, box in enumerate(expected_npboxes):
for pruned_box_ind, pruned_box in enumerate(pruned_npboxes):
if np.array_equal(pruned_box, box):
pruned_inds[box_ind] = pruned_box_ind
self.assertTrue(np.all(pruned_inds is not None))
expected_labels = ObjectDetectionLabels(
expected_npboxes[pruned_inds],
expected_class_ids[pruned_inds],
scores=expected_scores[pruned_inds])
pruned_labels.assert_equal(expected_labels)
class ObjectDetectionLabelsTest(unittest.TestCase):
def setUp(self):
self.class_map = ClassMap([ClassItem(1, 'car'), ClassItem(2, 'house')])
self.npboxes = np.array([
[0., 0., 2., 2.],
[2., 2., 4., 4.],
])
self.class_ids = np.array([1, 2])
self.scores = np.array([0.9, 0.9])
self.labels = ObjectDetectionLabels(
self.npboxes, self.class_ids, scores=self.scores)
def test_from_boxlist(self):
from object_detection.utils.np_box_list import BoxList
boxlist = BoxList(self.npboxes)
boxlist.add_field('classes', self.class_ids)
boxlist.add_field('scores', self.scores)
labels = ObjectDetectionLabels.from_boxlist(boxlist)
labels.assert_equal(self.labels)
def test_make_empty(self):
npboxes = np.empty((0, 4))
class_ids = np.empty((0, ))
scores = np.empty((0, ))
expected_labels = ObjectDetectionLabels(
npboxes, class_ids, scores=scores)
labels = ObjectDetectionLabels.make_empty()
labels.assert_equal(expected_labels)
def test_constructor(self):
labels = ObjectDetectionLabels(
self.npboxes, self.class_ids, scores=self.scores)
expected_labels = ObjectDetectionLabels(self.npboxes, self.class_ids,
self.scores)
labels.assert_equal(expected_labels)
labels = ObjectDetectionLabels(self.npboxes, self.class_ids)
scores = np.ones(self.class_ids.shape)
expected_labels = ObjectDetectionLabels(
self.npboxes, self.class_ids, scores=scores)
labels.assert_equal(expected_labels)
def test_get_boxes(self):
boxes = self.labels.get_boxes()
self.assertEqual(len(boxes), 2)
np.testing.assert_array_equal(boxes[0].npbox_format(),
self.npboxes[0, :])
np.testing.assert_array_equal(boxes[1].npbox_format(),
self.npboxes[1, :])
def test_len(self):
nb_labels = len(self.labels)
self.assertEqual(self.npboxes.shape[0], nb_labels)
def test_local_to_global(self):
local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
window = Box.make_square(10, 10, 10)
global_npboxes = ObjectDetectionLabels.local_to_global(
local_npboxes, window)
expected_global_npboxes = np.array([[10., 10., 12., 12.],
[12., 12., 14., 14.]])
np.testing.assert_array_equal(global_npboxes, expected_global_npboxes)
def test_global_to_local(self):
global_npboxes = np.array([[10., 10., 12., 12.], [12., 12., 14., 14.]])
window = Box.make_square(10, 10, 10)
local_npboxes = ObjectDetectionLabels.global_to_local(
global_npboxes, window)
expected_local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
np.testing.assert_array_equal(local_npboxes, expected_local_npboxes)
def test_local_to_normalized(self):
local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
window = Box(0, 0, 10, 100)
norm_npboxes = ObjectDetectionLabels.local_to_normalized(
local_npboxes, window)
expected_norm_npboxes = np.array([[0., 0., 0.2, 0.02],
[0.2, 0.02, 0.4, 0.04]])
np.testing.assert_array_equal(norm_npboxes, expected_norm_npboxes)
def test_normalized_to_local(self):
norm_npboxes = np.array([[0., 0., 0.2, 0.02], [0.2, 0.02, 0.4, 0.04]])
window = Box(0, 0, 10, 100)
local_npboxes = ObjectDetectionLabels.normalized_to_local(
norm_npboxes, window)
expected_local_npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.]])
np.testing.assert_array_equal(local_npboxes, expected_local_npboxes)
def test_get_overlapping(self):
window = Box.make_square(0, 0, 2.01)
labels = ObjectDetectionLabels.get_overlapping(self.labels, window)
labels.assert_equal(self.labels)
window = Box.make_square(0, 0, 3)
labels = ObjectDetectionLabels.get_overlapping(
self.labels, window, ioa_thresh=0.5)
npboxes = np.array([[0., 0., 2., 2.]])
class_ids = np.array([1])
scores = np.array([0.9])
expected_labels = ObjectDetectionLabels(
npboxes, class_ids, scores=scores)
labels.assert_equal(expected_labels)
window = Box.make_square(0, 0, 3)
labels = ObjectDetectionLabels.get_overlapping(
self.labels, window, ioa_thresh=0.1, clip=True)
expected_npboxes = np.array([
[0., 0., 2., 2.],
[2., 2., 3., 3.],
])
expected_labels = ObjectDetectionLabels(
expected_npboxes, self.class_ids, scores=self.scores)
labels.assert_equal(expected_labels)
def test_concatenate(self):
npboxes = np.array([[4., 4., 5., 5.]])
class_ids = np.array([2])
scores = np.array([0.3])
labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
new_labels = ObjectDetectionLabels.concatenate(self.labels, labels)
npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.],
[4., 4., 5., 5.]])
class_ids = np.array([1, 2, 2])
scores = np.array([0.9, 0.9, 0.3])
expected_labels = ObjectDetectionLabels(
npboxes, class_ids, scores=scores)
new_labels.assert_equal(expected_labels)
def test_prune_duplicates(self):
# This first box has a score below score_thresh so it should get
# pruned. The third box overlaps with the second, but has higher score,
# so the second one should get pruned. The fourth box overlaps with
# the second less than merge_thresh, so it should not get pruned.
npboxes = np.array([[0., 0., 2., 2.], [2., 2., 4., 4.],
[2.1, 2.1, 4.1, 4.1], [3.5, 3.5, 5.5, 5.5]])
class_ids = np.array([1, 2, 1, 2])
scores = np.array([0.2, 0.9, 0.9, 1.0])
labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
score_thresh = 0.5
merge_thresh = 0.5
pruned_labels = ObjectDetectionLabels.prune_duplicates(
labels, score_thresh, merge_thresh)
self.assertEqual(len(pruned_labels), 2)
expected_npboxes = np.array([[2.1, 2.1, 4.1, 4.1],
[3.5, 3.5, 5.5, 5.5]])
expected_class_ids = np.array([1, 2])
expected_scores = np.array([0.9, 1.0])
# prune_duplicates does not maintain ordering of boxes, so find match
# between pruned boxes and expected_npboxes.
pruned_npboxes = pruned_labels.get_npboxes()
pruned_inds = [None, None]
for box_ind, box in enumerate(expected_npboxes):
for pruned_box_ind, pruned_box in enumerate(pruned_npboxes):
if np.array_equal(pruned_box, box):
pruned_inds[box_ind] = pruned_box_ind
self.assertTrue(np.all(pruned_inds is not None))
expected_labels = ObjectDetectionLabels(
expected_npboxes[pruned_inds],
expected_class_ids[pruned_inds],
scores=expected_scores[pruned_inds])
pruned_labels.assert_equal(expected_labels)
def test_filter_by_aoi(self):
aois = [Box.make_square(0, 0, 2).to_shapely()]
filt_labels = self.labels.filter_by_aoi(aois)
npboxes = np.array([[0., 0., 2., 2.]])
class_ids = np.array([1])
scores = np.array([0.9])
exp_labels = ObjectDetectionLabels(npboxes, class_ids, scores=scores)
self.assertEqual(filt_labels, exp_labels)
aois = [Box.make_square(4, 4, 2).to_shapely()]
filt_labels = self.labels.filter_by_aoi(aois)
exp_labels = ObjectDetectionLabels.make_empty()
self.assertEqual(filt_labels, exp_labels)