def test_raise_error_on_empty_config(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
with self.assertRaises(ValueError):
losses_builder.build(losses_proto)
def test_build_all_loss_parameters(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
classification_loss {
weighted_softmax {
}
}
hard_example_miner {
}
classification_weight: 0.8
localization_weight: 0.2
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
(classification_loss, localization_loss, classification_weight,
localization_weight,
hard_example_miner) = losses_builder.build(losses_proto)
self.assertTrue(isinstance(hard_example_miner,
losses.HardExampleMiner))
self.assertTrue(
isinstance(classification_loss,
losses.WeightedSoftmaxClassificationLoss))
self.assertTrue(
isinstance(localization_loss, losses.WeightedL2LocalizationLoss))
self.assertAlmostEqual(classification_weight, 0.8)
self.assertAlmostEqual(localization_weight, 0.2)
def test_build_hard_example_miner_with_non_default_values(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
classification_loss {
weighted_softmax {
}
}
hard_example_miner {
num_hard_examples: 32
iou_threshold: 0.5
loss_type: LOCALIZATION
max_negatives_per_positive: 10
min_negatives_per_image: 3
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
_, _, _, _, hard_example_miner = losses_builder.build(losses_proto)
self.assertTrue(isinstance(hard_example_miner,
losses.HardExampleMiner))
self.assertEqual(hard_example_miner._num_hard_examples, 32)
self.assertAlmostEqual(hard_example_miner._iou_threshold, 0.5)
self.assertEqual(hard_example_miner._max_negatives_per_positive, 10)
self.assertEqual(hard_example_miner._min_negatives_per_image, 3)
def test_raise_error_when_both_focal_loss_and_hard_example_miner(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
classification_loss {
weighted_sigmoid_focal {
}
}
hard_example_miner {
}
classification_weight: 0.8
localization_weight: 0.2
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
with self.assertRaises(ValueError):
losses_builder.build(losses_proto)
def _build_ssd_model(ssd_config, is_training):
"""Builds an SSD detection model based on the model config.
Args:
ssd_config: A ssd.proto object containing the config for the desired
SSDMetaArch.
is_training: True if this model is being built for training purposes.
Returns:
SSDMetaArch based on the config.
Raises:
ValueError: If ssd_config.type is not recognized (i.e. not registered in
model_class_map).
"""
num_classes = ssd_config.num_classes
# Feature extractor
feature_extractor = _build_ssd_feature_extractor(ssd_config.feature_extractor,
is_training)
box_coder = box_coder_builder.build(ssd_config.box_coder)
matcher = matcher_builder.build(ssd_config.matcher)
region_similarity_calculator = sim_calc.build(
ssd_config.similarity_calculator)
ssd_box_predictor = box_predictor_builder.build(hyperparams_builder.build,
ssd_config.box_predictor,
is_training, num_classes)
anchor_generator = anchor_generator_builder.build(
ssd_config.anchor_generator)
image_resizer_fn = image_resizer_builder.build(ssd_config.image_resizer)
non_max_suppression_fn, score_conversion_fn = post_processing_builder.build(
ssd_config.post_processing)
(classification_loss, localization_loss, classification_weight,
localization_weight,
hard_example_miner) = losses_builder.build(ssd_config.loss)
normalize_loss_by_num_matches = ssd_config.normalize_loss_by_num_matches
return ssd_meta_arch.SSDMetaArch(
is_training,
anchor_generator,
ssd_box_predictor,
box_coder,
feature_extractor,
matcher,
region_similarity_calculator,
image_resizer_fn,
non_max_suppression_fn,
score_conversion_fn,
classification_loss,
localization_loss,
classification_weight,
localization_weight,
normalize_loss_by_num_matches,
hard_example_miner)
def test_do_not_build_hard_example_miner_by_default(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
classification_loss {
weighted_softmax {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
_, _, _, _, hard_example_miner = losses_builder.build(losses_proto)
self.assertEqual(hard_example_miner, None)
def test_build_weighted_iou_localization_loss(self):
losses_text_proto = """
localization_loss {
weighted_iou {
}
}
classification_loss {
weighted_softmax {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
_, localization_loss, _, _, _ = losses_builder.build(losses_proto)
self.assertTrue(
isinstance(localization_loss, losses.WeightedIOULocalizationLoss))
def test_build_weighted_sigmoid_classification_loss(self):
losses_text_proto = """
classification_loss {
weighted_sigmoid {
}
}
localization_loss {
weighted_l2 {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
classification_loss, _, _, _, _ = losses_builder.build(losses_proto)
self.assertTrue(
isinstance(classification_loss,
losses.WeightedSigmoidClassificationLoss))
def test_build_weighted_softmax_classification_loss_with_logit_scale(self):
losses_text_proto = """
classification_loss {
weighted_softmax {
logit_scale: 2.0
}
}
localization_loss {
weighted_l2 {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
classification_loss, _, _, _, _ = losses_builder.build(losses_proto)
self.assertTrue(
isinstance(classification_loss,
losses.WeightedSoftmaxClassificationLoss))
def test_build_weighted_sigmoid_focal_classification_loss(self):
losses_text_proto = """
classification_loss {
weighted_sigmoid_focal {
}
}
localization_loss {
weighted_l2 {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
classification_loss, _, _, _, _ = losses_builder.build(losses_proto)
self.assertTrue(
isinstance(classification_loss,
losses.SigmoidFocalClassificationLoss))
self.assertAlmostEqual(classification_loss._alpha, None)
self.assertAlmostEqual(classification_loss._gamma, 2.0)
def test_build_hard_example_miner_for_localization_loss(self):
losses_text_proto = """
localization_loss {
weighted_l2 {
}
}
classification_loss {
weighted_softmax {
}
}
hard_example_miner {
loss_type: LOCALIZATION
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
_, _, _, _, hard_example_miner = losses_builder.build(losses_proto)
self.assertTrue(isinstance(hard_example_miner,
losses.HardExampleMiner))
self.assertEqual(hard_example_miner._loss_type, 'loc')
def test_anchorwise_output(self):
losses_text_proto = """
classification_loss {
weighted_sigmoid {
anchorwise_output: true
}
}
localization_loss {
weighted_l2 {
}
}
"""
losses_proto = losses_pb2.Loss()
text_format.Merge(losses_text_proto, losses_proto)
classification_loss, _, _, _, _ = losses_builder.build(losses_proto)
self.assertTrue(
isinstance(classification_loss,
losses.WeightedSigmoidClassificationLoss))
predictions = tf.constant([[[0.0, 1.0, 0.0], [0.0, 0.5, 0.5]]])
targets = tf.constant([[[0.0, 1.0, 0.0], [0.0, 0.0, 1.0]]])
weights = tf.constant([[1.0, 1.0]])
loss = classification_loss(predictions, targets, weights=weights)
self.assertEqual(loss.shape, [1, 2])