def test_single_class_foreground_with_tiny_iou_threshold_top_2():
scores_pred = tf.constant([[0.4, 0.6], [0.5, 0.5], [0.1, 0.9], [0.3, 0.7],
[0.2, 0.8]])
batched_scores_pred = tf.expand_dims(scores_pred, axis=0)
# scores in descending order
# score 0.9 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked
# score 0.8 -- bbox [0, 1, 1, 2] -- should not be picked since its iou with first box is 1/7
# score 0.7 -- bbox [0, 0, 1, 1] -- should not be picked since its iou with first box is 1/7, with second box is 0
# score 0.6 -- bbox [-0.2, 1, 0.8, 2] -- should not be picked since its iou with first box is 0.15 / (2. - 0.15)
# score 0.5 -- bbox [-2, -2, -1, -1] -- should be picked since since it doesn't intersect with any other box
boxes_pred = tf.constant([
[-0.2, 1.0, 0.8, 2.0],
[-2.0, -2.0, -1.0, -1.0],
[0.5, 0.5, 1.5, 1.5],
[0.0, 0.0, 1.0, 1.0],
[0.0, 1.0, 1.0, 2.0],
])
batched_boxes_pred = tf.expand_dims(boxes_pred, axis=0)
decoder = DetectionDecoder(score_threshold=0.0,
iou_threshold=0.15 / 1.85,
top_k=2,
max_nms_size=5)
top_k_scores, top_k_boxes = decoder(batched_scores_pred,
batched_boxes_pred)
expected_top_k_scores = (np.asarray([[1.0, 0.9],
[1.0,
0.5]]).astype(np.float32).reshape(
(1, 2, 2)))
expected_top_k_boxes = (np.asarray([[0.5, 0.5, 1.5, 1.5],
[-2.0, -2.0, -1.0,
-1.0]]).astype(np.float32).reshape(
(1, 2, 4)))
np.testing.assert_allclose(expected_top_k_scores, top_k_scores)
np.testing.assert_allclose(expected_top_k_boxes, top_k_boxes)
def test_two_classes_foreground_with_larger_iou_threshold_top_5():
scores_pred = tf.constant([
[0.4, 0.6, 0.9],
[0.5, 0.5, 0.7],
[0.1, 0.9, 0.5],
[0.3, 0.7, 0.2],
[0.2, 0.8, 1.2],
])
batched_scores_pred = tf.expand_dims(scores_pred, axis=0)
# class 1
# scores in descending order
# score 0.9 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked and compared with 2nd class
# score 0.8 -- bbox [0, 1, 1, 2] -- should be picked and compared with 2nd class
# score 0.7 -- bbox [0, 0, 1, 1] -- should be picked and compared with 2nd class
# score 0.6 -- bbox [-0.2, 1, 0.8, 2] -- should not be picked since max_nms_size is 3
# score 0.5 -- bbox [-2, -2, -1, -1] -- should not be picked since max_nms_size is 3
# class 2
# scores in descending order
# score 1.2 -- bbox [0, 1, 1, 2] -- should be picked and compared with 1st class
# score 0.9 -- bbox [-0.2, 1., 0.8, 2.] -- should be picked and compared since its iou is 2/3 with first bbox
# score 0.7 -- bbox [-2, -2, -1, -1] -- should be picked and compared with 1st class
# score 0.5 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked and compared since its iou is 1/7 with first bbox
# score 0.2 -- bbox [0, 0, 1, 1] -- should not be picked since max_nms_size is 3
boxes_pred = tf.constant([
[-0.2, 1.0, 0.8, 2.0],
[-2.0, -2.0, -1.0, -1.0],
[0.5, 0.5, 1.5, 1.5],
[0.0, 0.0, 1.0, 1.0],
[0.0, 1.0, 1.0, 2.0],
])
batched_boxes_pred = tf.expand_dims(boxes_pred, axis=0)
decoder = DetectionDecoder(score_threshold=0.0,
iou_threshold=0.7,
top_k=5,
max_nms_size=3)
top_k_scores, top_k_boxes = decoder(batched_scores_pred,
batched_boxes_pred)
expected_top_k_scores = (np.asarray([[2.0, 1.2], [1.0, 0.9], [2.0, 0.9],
[1.0, 0.8],
[1.0,
0.7]]).astype(np.float32).reshape(
(1, 5, 2)))
expected_top_k_boxes = (np.asarray([
[0.0, 1.0, 1.0, 2.0],
[0.5, 0.5, 1.5, 1.5],
[-0.2, 1.0, 0.8, 2.0],
[0.0, 1.0, 1.0, 2.0],
[0.0, 0.0, 1.0, 1.0],
]).astype(np.float32).reshape((1, 5, 4)))
np.testing.assert_allclose(expected_top_k_scores, top_k_scores)
np.testing.assert_allclose(expected_top_k_boxes, top_k_boxes)
def test_single_class_batched_foreground_with_large_iou_threshold():
batched_scores_pred = tf.constant([
[[0.4, 0.6], [0.5, 0.5], [0.1, 0.9], [0.3, 0.7], [0.2, 0.8]],
[[0.4, 0.9], [0.5, 0.7], [0.1, 0.5], [0.3, 0.2], [0.2, 1.2]],
])
# batch 1
# scores in descending order
# score 0.9 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked
# score 0.8 -- bbox [0, 1, 1, 2] -- should not be picked
# score 0.7 -- bbox [0, 0, 1, 1] -- should be picked
# score 0.6 -- bbox [-0.2, 1, 0.8, 2] -- should be picked
# score 0.5 -- bbox [-2, -2, -1, -1] -- should not be picked
# batch 2
# scores in descending order
# score 1.2 -- bbox [0, 1, 1, 2] -- should be picked
# score 0.9 -- bbox [-0.2, 1., 0.8, 2.] -- should not be picked since its iou is 2/3 with first bbox
# score 0.7 -- bbox [-2, -2, -1, -1] -- should be picked
# score 0.5 -- bbox [0.5, 0.5, 1.5, 1.5] -- should not be picked since its iou is 1/7 with first bbox
# score 0.2 -- bbox [0, 0, 1, 1] -- should not be picked
boxes_pred = tf.constant([
[-0.2, 1.0, 0.8, 2.0],
[-2.0, -2.0, -1.0, -1.0],
[0.5, 0.5, 1.5, 1.5],
[0.0, 0.0, 1.0, 1.0],
[0.0, 1.0, 1.0, 2.0],
])
batched_boxes_pred = tf.tile(tf.expand_dims(boxes_pred, axis=0), [2, 1, 1])
decoder = DetectionDecoder(score_threshold=0.0,
iou_threshold=1 / 7,
top_k=3,
max_nms_size=5)
top_k_scores, top_k_boxes = decoder(batched_scores_pred,
batched_boxes_pred)
print("top k scores {}".format(top_k_scores))
expected_top_k_scores = np.asarray([[[1.0, 0.9], [1.0, 0.6], [1.0, 0.5]],
[[1.0, 1.2], [1.0, 0.7],
[1.0, 0.2]]]).astype(np.float32)
expected_top_k_boxes = np.asarray([
[[0.5, 0.5, 1.5, 1.5], [-0.2, 1, 0.8, 2], [-2, -2, -1, -1]],
[[0, 1, 1, 2], [-2, -2, -1, -1], [0, 0, 1, 1]],
]).astype(np.float32)
np.testing.assert_allclose(expected_top_k_scores, top_k_scores)
np.testing.assert_allclose(expected_top_k_boxes, top_k_boxes)
def test_single_class_foreground_with_large_iou_threshold_top_5_normalized():
scores_pred = tf.constant([[0.4, 0.6], [0.5, 0.5], [0.1, 0.9], [0.3, 0.7],
[0.2, 0.8]])
batched_scores_pred = tf.expand_dims(scores_pred, axis=0)
# scores in descending order
# score 0.9 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked
# score 0.8 -- bbox [0, 1, 1, 2] -- should be picked since its iou with first box is 1/7
# score 0.7 -- bbox [0, 0, 1, 1] -- should be picked since its iou with first box is 1/7, with second box is 0
# score 0.6 -- bbox [-0.2, 1, 0.8, 2] -- should not be picked since its iou with 2nd box is 0.66
# score 0.5 -- bbox [-2, -2, -1, -1] -- should be picked
# all normaliezd by 2.
boxes_pred = tf.constant([
[-0.1, 0.5, 0.4, 1.0],
[-1.0, -1.0, -0.5, -0.5],
[0.25, 0.25, 0.75, 0.75],
[0.0, 0.0, 0.5, 0.5],
[0.0, 0.5, 0.5, 1.0],
])
batched_boxes_pred = tf.expand_dims(boxes_pred, axis=0)
decoder = DetectionDecoder(score_threshold=0.0,
iou_threshold=0.25,
top_k=5,
max_nms_size=5)
top_k_scores, top_k_boxes = decoder(batched_scores_pred,
batched_boxes_pred)
expected_top_k_scores = (np.asarray([[1.0, 0.9], [1.0, 0.8], [1.0, 0.7],
[1.0, 0.5],
[0.0,
0.0]]).astype(np.float32).reshape(
(1, 5, 2)))
expected_top_k_boxes = (np.asarray([
[0.25, 0.25, 0.75, 0.75],
[0.0, 0.5, 0.5, 1.0],
[0.0, 0.0, 0.5, 0.5],
[-1.0, -1.0, -0.5, -0.5],
[0.0, 0.0, 0.0, 0.0],
]).astype(np.float32).reshape((1, 5, 4)))
np.testing.assert_allclose(expected_top_k_scores, top_k_scores)
np.testing.assert_allclose(expected_top_k_boxes, top_k_boxes)
def test_single_class_foreground_with_large_iou_threshold_nms_2():
scores_pred = tf.constant([[0.4, 0.6], [0.5, 0.5], [0.1, 0.9], [0.3, 0.7],
[0.2, 0.8]])
batched_scores_pred = tf.expand_dims(scores_pred, axis=0)
# scores in descending order
# score 0.9 -- bbox [0.5, 0.5, 1.5, 1.5] -- should be picked
# score 0.8 -- bbox [0, 1, 1, 2] -- should be picked since its iou with first box is 1/7
# score 0.7 -- bbox [0, 0, 1, 1] -- should not be picked since max_nms_size is 2
# score 0.6 -- bbox [-0.2, 1, 0.8, 2] -- should not be picked
# score 0.5 -- bbox [-2, -2, -1, -1] -- should not be picked
boxes_pred = tf.constant([
[-0.2, 1.0, 0.8, 2.0],
[-2.0, -2.0, -1.0, -1.0],
[0.5, 0.5, 1.5, 1.5],
[0.0, 0.0, 1.0, 1.0],
[0.0, 1.0, 1.0, 2.0],
])
batched_boxes_pred = tf.expand_dims(boxes_pred, axis=0)
decoder = DetectionDecoder(score_threshold=0.0,
iou_threshold=0.25,
top_k=3,
max_nms_size=2)
top_k_scores, top_k_boxes = decoder(batched_scores_pred,
batched_boxes_pred)
# Since top_k > max_nms_size, the last one would be padded zero.
expected_top_k_scores = (np.asarray([[1.0, 0.9], [1.0, 0.8],
[0.0,
0.0]]).astype(np.float32).reshape(
(1, 3, 2)))
expected_top_k_boxes = (np.asarray([[0.5, 0.5, 1.5, 1.5],
[0.0, 1.0, 1.0, 2.0],
[0.0, 0.0, 0.0,
0.0]]).astype(np.float32).reshape(
(1, 3, 4)))
np.testing.assert_allclose(expected_top_k_scores, top_k_scores)
np.testing.assert_allclose(expected_top_k_boxes, top_k_boxes)