def test_first_rows_close_as_set(self):
a = [1, 2, 3, 0, 0]
b = [3, 2, 1, 0, 0]
k = 3
self.assertTrue(test_utils.first_rows_close_as_set(a, b, k))
a = [[1, 2], [1, 4], [0, 0]]
b = [[1, 4 + 1e-9], [1, 2], [0, 0]]
k = 2
self.assertTrue(test_utils.first_rows_close_as_set(a, b, k))
a = [[1, 2], [1, 4], [0, 0]]
b = [[1, 4 + 1e-9], [2, 2], [0, 0]]
k = 2
self.assertFalse(test_utils.first_rows_close_as_set(a, b, k))
def test_first_rows_close_as_set(self):
a = [1, 2, 3, 0, 0]
b = [3, 2, 1, 0, 0]
k = 3
self.assertTrue(test_utils.first_rows_close_as_set(a, b, k))
a = [[1, 2], [1, 4], [0, 0]]
b = [[1, 4 + 1e-9], [1, 2], [0, 0]]
k = 2
self.assertTrue(test_utils.first_rows_close_as_set(a, b, k))
a = [[1, 2], [1, 4], [0, 0]]
b = [[1, 4 + 1e-9], [2, 2], [0, 0]]
k = 2
self.assertFalse(test_utils.first_rows_close_as_set(a, b, k))
def test_postprocess_results_are_correct_static(self, use_keras):
with tf.Graph().as_default():
_, _, _, _ = self._create_model(use_keras=use_keras)
def graph_fn(input_image):
model, _, _, _ = self._create_model(use_static_shapes=True,
nms_max_size_per_class=4)
preprocessed_inputs, true_image_shapes = model.preprocess(input_image)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
return (detections['detection_boxes'], detections['detection_scores'],
detections['detection_classes'], detections['num_detections'],
detections['detection_multiclass_scores'])
batch_size = 2
image_size = 2
channels = 3
input_image = np.random.rand(batch_size, image_size, image_size,
channels).astype(np.float32)
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0]
]
] # padding
expected_scores = [[0, 0, 0, 0], [0, 0, 0, 0]]
expected_multiclass_scores = [[[0, 0], [0, 0], [0, 0], [0, 0]],
[[0, 0], [0, 0], [0, 0], [0, 0]]]
expected_classes = [[0, 0, 0, 0], [0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
(detection_boxes, detection_scores, detection_classes,
num_detections, detection_multiclass_scores) = self.execute(graph_fn,
[input_image])
for image_idx in range(batch_size):
self.assertTrue(test_utils.first_rows_close_as_set(
detection_boxes[image_idx][
0:expected_num_detections[image_idx]].tolist(),
expected_boxes[image_idx][0:expected_num_detections[image_idx]]))
self.assertAllClose(
detection_scores[image_idx][0:expected_num_detections[image_idx]],
expected_scores[image_idx][0:expected_num_detections[image_idx]])
self.assertAllClose(
detection_multiclass_scores[image_idx]
[0:expected_num_detections[image_idx]],
expected_multiclass_scores[image_idx]
[0:expected_num_detections[image_idx]])
self.assertAllClose(
detection_classes[image_idx][0:expected_num_detections[image_idx]],
expected_classes[image_idx][0:expected_num_detections[image_idx]])
self.assertAllClose(num_detections,
expected_num_detections)
def test_postprocess_results_are_correct(self, use_keras):
batch_size = 2
image_size = 2
input_shapes = [(batch_size, image_size, image_size, 3),
(None, image_size, image_size, 3),
(batch_size, None, None, 3),
(None, None, None, 3)]
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
]
] # padding
expected_scores = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_classes = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
for input_shape in input_shapes:
tf_graph = tf.Graph()
with tf_graph.as_default():
model, _, _, _ = self._create_model(use_keras=use_keras)
input_placeholder = tf.placeholder(tf.float32, shape=input_shape)
preprocessed_inputs, true_image_shapes = model.preprocess(
input_placeholder)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
self.assertIn('detection_boxes', detections)
self.assertIn('detection_scores', detections)
self.assertIn('detection_classes', detections)
self.assertIn('num_detections', detections)
init_op = tf.global_variables_initializer()
with self.test_session(graph=tf_graph) as sess:
sess.run(init_op)
detections_out = sess.run(detections,
feed_dict={
input_placeholder:
np.random.uniform(
size=(batch_size, 2, 2, 3))})
for image_idx in range(batch_size):
self.assertTrue(
test_utils.first_rows_close_as_set(
detections_out['detection_boxes'][image_idx].tolist(),
expected_boxes[image_idx]))
self.assertAllClose(detections_out['detection_scores'], expected_scores)
self.assertAllClose(detections_out['detection_classes'], expected_classes)
self.assertAllClose(detections_out['num_detections'],
expected_num_detections)
def test_postprocess_results_are_correct(self, use_keras):
batch_size = 2
image_size = 2
input_shapes = [(batch_size, image_size, image_size, 3),
(None, image_size, image_size, 3),
(batch_size, None, None, 3),
(None, None, None, 3)]
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
]
] # padding
expected_scores = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_classes = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
for input_shape in input_shapes:
tf_graph = tf.Graph()
with tf_graph.as_default():
model, _, _, _ = self._create_model(use_keras=use_keras)
input_placeholder = tf.placeholder(tf.float32, shape=input_shape)
preprocessed_inputs, true_image_shapes = model.preprocess(
input_placeholder)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
self.assertIn('detection_boxes', detections)
self.assertIn('detection_scores', detections)
self.assertIn('detection_classes', detections)
self.assertIn('num_detections', detections)
init_op = tf.global_variables_initializer()
with self.test_session(graph=tf_graph) as sess:
sess.run(init_op)
detections_out = sess.run(detections,
feed_dict={
input_placeholder:
np.random.uniform(
size=(batch_size, 2, 2, 3))})
for image_idx in range(batch_size):
self.assertTrue(
test_utils.first_rows_close_as_set(
detections_out['detection_boxes'][image_idx].tolist(),
expected_boxes[image_idx]))
self.assertAllClose(detections_out['detection_scores'], expected_scores)
self.assertAllClose(detections_out['detection_classes'], expected_classes)
self.assertAllClose(detections_out['num_detections'],
expected_num_detections)
def test_postprocess_results_are_correct_static(self, use_keras):
with tf.Graph().as_default():
_, _, _, _ = self._create_model(use_keras=use_keras)
def graph_fn(input_image):
model, _, _, _ = self._create_model(use_static_shapes=True,
nms_max_size_per_class=4)
preprocessed_inputs, true_image_shapes = model.preprocess(input_image)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
return (detections['detection_boxes'], detections['detection_scores'],
detections['detection_classes'], detections['num_detections'])
batch_size = 2
image_size = 2
channels = 3
input_image = np.random.rand(batch_size, image_size, image_size,
channels).astype(np.float32)
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0]
]
] # padding
expected_scores = [[0, 0, 0, 0], [0, 0, 0, 0]]
expected_classes = [[0, 0, 0, 0], [0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
(detection_boxes, detection_scores, detection_classes,
num_detections) = self.execute(graph_fn, [input_image])
for image_idx in range(batch_size):
self.assertTrue(test_utils.first_rows_close_as_set(
detection_boxes[image_idx][
0:expected_num_detections[image_idx]].tolist(),
expected_boxes[image_idx][0:expected_num_detections[image_idx]]))
self.assertAllClose(
detection_scores[image_idx][0:expected_num_detections[image_idx]],
expected_scores[image_idx][0:expected_num_detections[image_idx]])
self.assertAllClose(
detection_classes[image_idx][0:expected_num_detections[image_idx]],
expected_classes[image_idx][0:expected_num_detections[image_idx]])
self.assertAllClose(num_detections,
expected_num_detections)
def test_postprocess_results_are_correct_with_calibration(self, use_keras):
batch_size = 2
image_size = 2
input_shapes = [(batch_size, image_size, image_size, 3),
(None, image_size, image_size, 3),
(batch_size, None, None, 3),
(None, None, None, 3)]
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
]
] # padding
# Calibration mapping value below is set to map all scores to 0.5, except
# for the last two detections in each batch (see expected number of
# detections below.
expected_scores = [[0.5, 0.5, 0.5, 0., 0.], [0.5, 0.5, 0.5, 0., 0.]]
expected_classes = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
raw_detection_boxes = [[[0., 0., 0.5, 0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5], [1., 1., 1.5, 1.5]],
[[0., 0., 0.5, 0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5], [1., 1., 1.5, 1.5]]]
raw_detection_scores = [[[0.5, 0.5], [0.5, 0.5], [0.5, 0.5], [0.5, 0.5]],
[[0.5, 0.5], [0.5, 0.5], [0.5, 0.5], [0.5, 0.5]]]
for input_shape in input_shapes:
tf_graph = tf.Graph()
with tf_graph.as_default():
model, _, _, _ = self._create_model(use_keras=use_keras,
calibration_mapping_value=0.5)
input_placeholder = tf.placeholder(tf.float32, shape=input_shape)
preprocessed_inputs, true_image_shapes = model.preprocess(
input_placeholder)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
self.assertIn('detection_boxes', detections)
self.assertIn('detection_scores', detections)
self.assertIn('detection_classes', detections)
self.assertIn('num_detections', detections)
self.assertIn('raw_detection_boxes', detections)
self.assertIn('raw_detection_scores', detections)
init_op = tf.global_variables_initializer()
with self.test_session(graph=tf_graph) as sess:
sess.run(init_op)
detections_out = sess.run(detections,
feed_dict={
input_placeholder:
np.random.uniform(
size=(batch_size, 2, 2, 3))})
for image_idx in range(batch_size):
self.assertTrue(
test_utils.first_rows_close_as_set(
detections_out['detection_boxes'][image_idx].tolist(),
expected_boxes[image_idx]))
self.assertAllClose(detections_out['detection_scores'], expected_scores)
self.assertAllClose(detections_out['detection_classes'], expected_classes)
self.assertAllClose(detections_out['num_detections'],
expected_num_detections)
self.assertAllEqual(detections_out['raw_detection_boxes'],
raw_detection_boxes)
self.assertAllEqual(detections_out['raw_detection_scores'],
raw_detection_scores)
def test_postprocess_results_are_correct(self):
with test_utils.GraphContextOrNone() as g:
model, _, _, _ = self._create_model()
def graph_fn():
size = tf.random.uniform([], minval=2, maxval=3, dtype=tf.int32)
batch = tf.random.uniform([], minval=2, maxval=3, dtype=tf.int32)
shape = tf.stack([batch, size, size, 3])
image = tf.random.uniform(shape)
preprocessed_inputs, true_image_shapes = model.preprocess(image)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
return [
batch, detections['detection_boxes'],
detections['detection_scores'],
detections['detection_classes'],
detections['detection_multiclass_scores'],
detections['num_detections'],
detections['raw_detection_boxes'],
detections['raw_detection_scores'],
detections['detection_anchor_indices']
]
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
]
] # padding
expected_scores = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_multiclass_scores = [[[0, 0], [0, 0], [0, 0], [0, 0], [0, 0]],
[[0, 0], [0, 0], [0, 0], [0, 0], [0, 0]]]
expected_classes = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
expected_raw_detection_boxes = [[[0., 0., 0.5,
0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5],
[1., 1., 1.5, 1.5]],
[[0., 0., 0.5,
0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5],
[1., 1., 1.5, 1.5]]]
expected_raw_detection_scores = [[[0, 0], [0, 0], [0, 0], [0, 0]],
[[0, 0], [0, 0], [0, 0], [0, 0]]]
expected_detection_anchor_indices = [[0, 1, 2], [0, 1, 2]]
(batch, detection_boxes, detection_scores, detection_classes,
detection_multiclass_scores, num_detections, raw_detection_boxes,
raw_detection_scores,
detection_anchor_indices) = self.execute_cpu(graph_fn, [], graph=g)
for image_idx in range(batch):
self.assertTrue(
test_utils.first_rows_close_as_set(
detection_boxes[image_idx].tolist(),
expected_boxes[image_idx]))
self.assertSameElements(
detection_anchor_indices[image_idx],
expected_detection_anchor_indices[image_idx])
self.assertAllClose(detection_scores, expected_scores)
self.assertAllClose(detection_classes, expected_classes)
self.assertAllClose(detection_multiclass_scores,
expected_multiclass_scores)
self.assertAllClose(num_detections, expected_num_detections)
self.assertAllEqual(raw_detection_boxes, expected_raw_detection_boxes)
self.assertAllEqual(raw_detection_scores,
expected_raw_detection_scores)
def test_postprocess_results_are_correct_with_calibration(self, use_keras):
batch_size = 2
image_size = 2
input_shapes = [(batch_size, image_size, image_size, 3),
(None, image_size, image_size, 3),
(batch_size, None, None, 3),
(None, None, None, 3)]
expected_boxes = [
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
], # padding
[
[0, 0, .5, .5],
[0, .5, .5, 1],
[.5, 0, 1, .5],
[0, 0, 0, 0], # pruned prediction
[0, 0, 0, 0]
]
] # padding
# Calibration mapping value below is set to map all scores to 0.5, except
# for the last two detections in each batch (see expected number of
# detections below.
expected_scores = [[0.5, 0.5, 0.5, 0., 0.], [0.5, 0.5, 0.5, 0., 0.]]
expected_classes = [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]
expected_num_detections = np.array([3, 3])
raw_detection_boxes = [[[0., 0., 0.5, 0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5], [1., 1., 1.5, 1.5]],
[[0., 0., 0.5, 0.5], [0., 0.5, 0.5, 1.],
[0.5, 0., 1., 0.5], [1., 1., 1.5, 1.5]]]
raw_detection_scores = [[[0, 0], [0, 0], [0, 0], [0, 0]],
[[0, 0], [0, 0], [0, 0], [0, 0]]]
for input_shape in input_shapes:
tf_graph = tf.Graph()
with tf_graph.as_default():
model, _, _, _ = self._create_model(use_keras=use_keras,
calibration_mapping_value=0.5)
input_placeholder = tf.placeholder(tf.float32, shape=input_shape)
preprocessed_inputs, true_image_shapes = model.preprocess(
input_placeholder)
prediction_dict = model.predict(preprocessed_inputs,
true_image_shapes)
detections = model.postprocess(prediction_dict, true_image_shapes)
self.assertIn('detection_boxes', detections)
self.assertIn('detection_scores', detections)
self.assertIn('detection_classes', detections)
self.assertIn('num_detections', detections)
self.assertIn('raw_detection_boxes', detections)
self.assertIn('raw_detection_scores', detections)
init_op = tf.global_variables_initializer()
with self.test_session(graph=tf_graph) as sess:
sess.run(init_op)
detections_out = sess.run(detections,
feed_dict={
input_placeholder:
np.random.uniform(
size=(batch_size, 2, 2, 3))})
for image_idx in range(batch_size):
self.assertTrue(
test_utils.first_rows_close_as_set(
detections_out['detection_boxes'][image_idx].tolist(),
expected_boxes[image_idx]))
self.assertAllClose(detections_out['detection_scores'], expected_scores)
self.assertAllClose(detections_out['detection_classes'], expected_classes)
self.assertAllClose(detections_out['num_detections'],
expected_num_detections)
self.assertAllEqual(detections_out['raw_detection_boxes'],
raw_detection_boxes)
self.assertAllEqual(detections_out['raw_detection_scores'],
raw_detection_scores)
