def _create_model(self,
apply_hard_mining=True,
normalize_loc_loss_by_codesize=False):
is_training = False
num_classes = 1
mock_anchor_generator = MockAnchorGenerator2x2()
mock_box_predictor = test_utils.MockBoxPredictor(
is_training, num_classes)
mock_box_coder = test_utils.MockBoxCoder()
fake_feature_extractor = FakeSSDFeatureExtractor()
mock_matcher = test_utils.MockMatcher()
region_similarity_calculator = sim_calc.IouSimilarity()
encode_background_as_zeros = False
def image_resizer_fn(image):
return [tf.identity(image), tf.shape(image)]
classification_loss = losses.WeightedSigmoidClassificationLoss()
localization_loss = losses.WeightedSmoothL1LocalizationLoss()
non_max_suppression_fn = functools.partial(
post_processing.batch_multiclass_non_max_suppression,
score_thresh=-20.0,
iou_thresh=1.0,
max_size_per_class=5,
max_total_size=5)
classification_loss_weight = 1.0
localization_loss_weight = 1.0
negative_class_weight = 1.0
normalize_loss_by_num_matches = False
hard_example_miner = None
if apply_hard_mining:
# This hard example miner is expected to be a no-op.
hard_example_miner = losses.HardExampleMiner(
num_hard_examples=None, iou_threshold=1.0)
code_size = 4
model = ssd_meta_arch.SSDMetaArch(
is_training,
mock_anchor_generator,
mock_box_predictor,
mock_box_coder,
fake_feature_extractor,
mock_matcher,
region_similarity_calculator,
encode_background_as_zeros,
negative_class_weight,
image_resizer_fn,
non_max_suppression_fn,
tf.identity,
classification_loss,
localization_loss,
classification_loss_weight,
localization_loss_weight,
normalize_loss_by_num_matches,
hard_example_miner,
add_summaries=False,
normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize)
return model, num_classes, mock_anchor_generator.num_anchors(
), code_size
def setUp(self):
"""Set up mock SSD model.
Here we set up a simple mock SSD model that will always predict 4
detections that happen to always be exactly the anchors that are set up
in the above MockAnchorGenerator. Because we let max_detections=5,
we will also always end up with an extra padded row in the detection
results.
"""
is_training = False
self._num_classes = 1
conv_hyperparams = {}
self._conv_hyperparams = conv_hyperparams
mock_anchor_generator = MockAnchorGenerator2x2()
mock_box_predictor = test_utils.MockBoxPredictor(
is_training, self._num_classes)
mock_class_predictor = test_utils.MockClassPredictor(
is_training, 7, self._conv_hyperparams, True, 0.5, 5, 0.0, False)
mock_box_coder = test_utils.MockBoxCoder()
fake_feature_extractor = FakeSSDFeatureExtractor()
mock_matcher = test_utils.MockMatcher()
region_similarity_calculator = sim_calc.IouSimilarity()
def image_resizer_fn(image):
return tf.identity(image)
classification_loss = losses.WeightedSigmoidClassificationLoss(
anchorwise_output=True)
localization_loss = losses.WeightedSmoothL1LocalizationLoss(
anchorwise_output=True)
classification_loss_in_image_level = losses.WeightedSigmoidClassificationLossInImageLevel(
)
non_max_suppression_fn = functools.partial(
post_processing.batch_multiclass_non_max_suppression,
score_thresh=-20.0,
iou_thresh=1.0,
max_size_per_class=5,
max_total_size=5)
classification_loss_weight = 1.0
localization_loss_weight = 1.0
classification_loss_in_image_level_weight = 1.0
normalize_loss_by_num_matches = False
# This hard example miner is expected to be a no-op.
hard_example_miner = losses.HardExampleMiner(num_hard_examples=None,
iou_threshold=1.0)
self._num_anchors = 4
self._code_size = 4
self._model = ssd_meta_arch.SSDMetaArch(
is_training, mock_anchor_generator, mock_box_predictor,
mock_class_predictor, mock_box_coder, fake_feature_extractor,
mock_matcher, region_similarity_calculator, image_resizer_fn,
non_max_suppression_fn, tf.identity, classification_loss,
localization_loss, classification_loss_in_image_level,
classification_loss_weight, localization_loss_weight,
classification_loss_in_image_level_weight,
normalize_loss_by_num_matches, hard_example_miner)
def _build_localization_loss(loss_config):
"""Builds a localization loss based on the loss config.
Args:
loss_config: A losses_pb2.LocalizationLoss object.
Returns:
Loss based on the config.
Raises:
ValueError: On invalid loss_config.
"""
if not isinstance(loss_config, losses_pb2.LocalizationLoss):
raise ValueError('loss_config not of type losses_pb2.LocalizationLoss.')
loss_type = loss_config.WhichOneof('localization_loss')
if loss_type == 'weighted_l2':
return losses.WeightedL2LocalizationLoss()
if loss_type == 'weighted_smooth_l1':
return losses.WeightedSmoothL1LocalizationLoss(
loss_config.weighted_smooth_l1.delta)
if loss_type == 'weighted_iou':
return losses.WeightedIOULocalizationLoss()
raise ValueError('Empty loss config.')
def build(self):
super(FasterRCNNFirstStageLoss, self).build()
self._proposal_target_assigner = (
target_assigner.create_target_assigner('FasterRCNN', 'proposal'))
self._sampler = (
balanced_positive_negative_sampler.BalancedPositiveNegativeSampler(
positive_fraction=self.positive_balance_fraction,
is_static=False))
self._localization_loss = od_losses.WeightedSmoothL1LocalizationLoss()
self._objectness_loss = od_losses.WeightedSoftmaxClassificationLoss()
return self
def testReturnsCorrectLoss(self):
batch_size = 2
num_anchors = 3
code_size = 4
prediction_tensor = tf.constant(
[[[2.5, 0, .4, 0], [0, 0, 0, 0], [0, 2.5, 0, .4]],
[[3.5, 0, 0, 0], [0, .4, 0, .9], [0, 0, 1.5, 0]]], tf.float32)
target_tensor = tf.zeros([batch_size, num_anchors, code_size])
weights = tf.constant([[2, 1, 1], [0, 3, 0]], tf.float32)
loss_op = losses.WeightedSmoothL1LocalizationLoss()
loss = loss_op(prediction_tensor, target_tensor, weights=weights)
exp_loss = 7.695
with self.test_session() as sess:
loss_output = sess.run(loss)
self.assertAllClose(loss_output, exp_loss)
def _build_classification_loss(loss_config):
"""Builds a classification loss based on the loss config.
Args:
loss_config: A losses_pb2.ClassificationLoss object.
Returns:
Loss based on the config.
Raises:
ValueError: On invalid loss_config.
"""
if not isinstance(loss_config, losses_pb2.ClassificationLoss):
raise ValueError(
'loss_config not of type losses_pb2.ClassificationLoss.')
loss_type = loss_config.WhichOneof('classification_loss')
if loss_type == 'weighted_sigmoid':
config = loss_config.weighted_sigmoid
return losses.WeightedSigmoidClassificationLoss(
anchorwise_output=config.anchorwise_output)
if loss_type == 'weighted_softmax':
config = loss_config.weighted_softmax
return losses.WeightedSoftmaxClassificationLoss(
anchorwise_output=config.anchorwise_output)
if loss_type == 'bootstrapped_sigmoid':
config = loss_config.bootstrapped_sigmoid
return losses.BootstrappedSigmoidClassificationLoss(
alpha=config.alpha,
bootstrap_type=('hard' if config.hard_bootstrap else 'soft'),
anchorwise_output=config.anchorwise_output)
if loss_type == 'weighted_l2':
config = loss_config.weighted_l2
return losses.WeightedL2LocalizationLoss(
anchorwise_output=config.anchorwise_output)
if loss_type == 'weighted_smooth_l1':
config = loss_config.weighted_smooth_l1
return losses.WeightedSmoothL1LocalizationLoss(
anchorwise_output=config.anchorwise_output)
raise ValueError('Empty loss config.')
def build_loss(loss_type):
"""Builds the desired type of loss
Args:
loss_type: loss type (e.g. 'berHu', 'smooth_l1')
Returns:
Class of the specified loss_type
"""
if loss_type == 'berHu':
return losses_custom.WeightedBerHu()
elif loss_type == 'chamfer_dist':
return losses_custom.ChamferDistance()
elif loss_type == 'emd':
return losses_custom.EarthMoversDistance()
elif loss_type == 'smooth_l1':
return losses.WeightedSmoothL1LocalizationLoss()
elif loss_type == 'smooth_l1_nonzero':
return losses_custom.WeightedNonZeroSmoothL1LocalizationLoss()
elif loss_type == 'softmax':
return losses.WeightedSoftmaxClassificationLoss()
elif loss_type == 'focal':
return losses.SigmoidFocalClassificationLoss()
elif loss_type == 'softmax_temp':
return losses.WeightedSoftmaxClassificationLoss(0.5)
elif loss_type == 'sigmoid_ce':
return losses_custom.SigmoidClassificationLoss()
else:
raise ValueError('Invalid loss type', loss_type)
def __init__(self):
super(FakeDetectionModel, self).__init__(num_classes=NUMBER_OF_CLASSES)
conv_hyperparams_text_proto = """
regularizer {
l1_regularizer {
weight: 0.0003
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.3
}
}
activation: RELU_6
"""
hyperparams_proto = hyperparams_pb2.Hyperparams()
text_format.Merge(conv_hyperparams_text_proto, hyperparams_proto)
self._class_predictor = class_predictor.ImageLevelConvolutionalClassPredictor(
is_training=False,
num_classes=NUMBER_OF_CLASSES_IN_IMAGE_LEVEL,
conv_hyperparams=hyperparams_proto,
use_dropout=True,
dropout_keep_prob=0.5,
kernel_size=5,
class_prediction_bias_init=0.0,
apply_sigmoid_to_scores=False)
self._classification_loss = losses.WeightedSigmoidClassificationLoss(
anchorwise_output=True)
self._localization_loss = losses.WeightedSmoothL1LocalizationLoss(
anchorwise_output=True)
self._classification_in_image_level_loss = losses.WeightedSigmoidClassificationLossInImageLevel(
)
def _create_model(self,
interleaved=False,
apply_hard_mining=True,
normalize_loc_loss_by_codesize=False,
add_background_class=True,
random_example_sampling=False,
use_expected_classification_loss_under_sampling=False,
min_num_negative_samples=1,
desired_negative_sampling_ratio=3,
unroll_length=1):
num_classes = NUM_CLASSES
is_training = False
mock_anchor_generator = MockAnchorGenerator2x2()
mock_box_predictor = test_utils.MockBoxPredictor(is_training, num_classes)
mock_box_coder = test_utils.MockBoxCoder()
if interleaved:
fake_feature_extractor = FakeLSTMInterleavedFeatureExtractor()
else:
fake_feature_extractor = FakeLSTMFeatureExtractor()
mock_matcher = test_utils.MockMatcher()
region_similarity_calculator = sim_calc.IouSimilarity()
encode_background_as_zeros = False
def image_resizer_fn(image):
return [tf.identity(image), tf.shape(image)]
classification_loss = losses.WeightedSigmoidClassificationLoss()
localization_loss = losses.WeightedSmoothL1LocalizationLoss()
non_max_suppression_fn = functools.partial(
post_processing.batch_multiclass_non_max_suppression,
score_thresh=-20.0,
iou_thresh=1.0,
max_size_per_class=5,
max_total_size=MAX_TOTAL_NUM_BOXES)
classification_loss_weight = 1.0
localization_loss_weight = 1.0
negative_class_weight = 1.0
normalize_loss_by_num_matches = False
hard_example_miner = None
if apply_hard_mining:
# This hard example miner is expected to be a no-op.
hard_example_miner = losses.HardExampleMiner(
num_hard_examples=None,
iou_threshold=1.0)
target_assigner_instance = target_assigner.TargetAssigner(
region_similarity_calculator,
mock_matcher,
mock_box_coder,
negative_class_weight=negative_class_weight)
code_size = 4
model = lstm_ssd_meta_arch.LSTMSSDMetaArch(
is_training=is_training,
anchor_generator=mock_anchor_generator,
box_predictor=mock_box_predictor,
box_coder=mock_box_coder,
feature_extractor=fake_feature_extractor,
encode_background_as_zeros=encode_background_as_zeros,
image_resizer_fn=image_resizer_fn,
non_max_suppression_fn=non_max_suppression_fn,
score_conversion_fn=tf.identity,
classification_loss=classification_loss,
localization_loss=localization_loss,
classification_loss_weight=classification_loss_weight,
localization_loss_weight=localization_loss_weight,
normalize_loss_by_num_matches=normalize_loss_by_num_matches,
hard_example_miner=hard_example_miner,
unroll_length=unroll_length,
target_assigner_instance=target_assigner_instance,
add_summaries=False)
return model, num_classes, mock_anchor_generator.num_anchors(), code_size
def __init__(self):
super(FakeDetectionModel, self).__init__(num_classes=NUMBER_OF_CLASSES)
self._classification_loss = losses.WeightedSigmoidClassificationLoss()
self._localization_loss = losses.WeightedSmoothL1LocalizationLoss()
def _create_model(
self,
model_fn=ssd_meta_arch.SSDMetaArch,
apply_hard_mining=True,
normalize_loc_loss_by_codesize=False,
add_background_class=True,
random_example_sampling=False,
expected_loss_weights=model_pb2.DetectionModel().ssd.loss.NONE,
min_num_negative_samples=1,
desired_negative_sampling_ratio=3,
use_keras=False,
predict_mask=False,
use_static_shapes=False,
nms_max_size_per_class=5,
calibration_mapping_value=None,
return_raw_detections_during_predict=False):
is_training = False
num_classes = 1
mock_anchor_generator = MockAnchorGenerator2x2()
if use_keras:
mock_box_predictor = test_utils.MockKerasBoxPredictor(
is_training, num_classes, add_background_class=add_background_class)
else:
mock_box_predictor = test_utils.MockBoxPredictor(
is_training, num_classes, add_background_class=add_background_class)
mock_box_coder = test_utils.MockBoxCoder()
if use_keras:
fake_feature_extractor = FakeSSDKerasFeatureExtractor()
else:
fake_feature_extractor = FakeSSDFeatureExtractor()
mock_matcher = test_utils.MockMatcher()
region_similarity_calculator = sim_calc.IouSimilarity()
encode_background_as_zeros = False
def image_resizer_fn(image):
return [tf.identity(image), tf.shape(image)]
classification_loss = losses.WeightedSigmoidClassificationLoss()
localization_loss = losses.WeightedSmoothL1LocalizationLoss()
non_max_suppression_fn = functools.partial(
post_processing.batch_multiclass_non_max_suppression,
score_thresh=-20.0,
iou_thresh=1.0,
max_size_per_class=nms_max_size_per_class,
max_total_size=nms_max_size_per_class,
use_static_shapes=use_static_shapes)
score_conversion_fn = tf.identity
calibration_config = calibration_pb2.CalibrationConfig()
if calibration_mapping_value:
calibration_text_proto = """
function_approximation {
x_y_pairs {
x_y_pair {
x: 0.0
y: %f
}
x_y_pair {
x: 1.0
y: %f
}}}""" % (calibration_mapping_value, calibration_mapping_value)
text_format.Merge(calibration_text_proto, calibration_config)
score_conversion_fn = (
post_processing_builder._build_calibrated_score_converter( # pylint: disable=protected-access
tf.identity, calibration_config))
classification_loss_weight = 1.0
localization_loss_weight = 1.0
negative_class_weight = 1.0
normalize_loss_by_num_matches = False
hard_example_miner = None
if apply_hard_mining:
# This hard example miner is expected to be a no-op.
hard_example_miner = losses.HardExampleMiner(
num_hard_examples=None, iou_threshold=1.0)
random_example_sampler = None
if random_example_sampling:
random_example_sampler = sampler.BalancedPositiveNegativeSampler(
positive_fraction=0.5)
target_assigner_instance = target_assigner.TargetAssigner(
region_similarity_calculator,
mock_matcher,
mock_box_coder,
negative_class_weight=negative_class_weight)
model_config = model_pb2.DetectionModel()
if expected_loss_weights == model_config.ssd.loss.NONE:
expected_loss_weights_fn = None
else:
raise ValueError('Not a valid value for expected_loss_weights.')
code_size = 4
kwargs = {}
if predict_mask:
kwargs.update({
'mask_prediction_fn': test_utils.MockMaskHead(num_classes=1).predict,
})
model = model_fn(
is_training=is_training,
anchor_generator=mock_anchor_generator,
box_predictor=mock_box_predictor,
box_coder=mock_box_coder,
feature_extractor=fake_feature_extractor,
encode_background_as_zeros=encode_background_as_zeros,
image_resizer_fn=image_resizer_fn,
non_max_suppression_fn=non_max_suppression_fn,
score_conversion_fn=score_conversion_fn,
classification_loss=classification_loss,
localization_loss=localization_loss,
classification_loss_weight=classification_loss_weight,
localization_loss_weight=localization_loss_weight,
normalize_loss_by_num_matches=normalize_loss_by_num_matches,
hard_example_miner=hard_example_miner,
target_assigner_instance=target_assigner_instance,
add_summaries=False,
normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize,
freeze_batchnorm=False,
inplace_batchnorm_update=False,
add_background_class=add_background_class,
random_example_sampler=random_example_sampler,
expected_loss_weights_fn=expected_loss_weights_fn,
return_raw_detections_during_predict=(
return_raw_detections_during_predict),
**kwargs)
return model, num_classes, mock_anchor_generator.num_anchors(), code_size
def _create_model(self,
model_fn=ssd_meta_arch.SSDMetaArch,
apply_hard_mining=True,
normalize_loc_loss_by_codesize=False,
add_background_class=True,
random_example_sampling=False,
weight_regression_loss_by_score=False,
use_expected_classification_loss_under_sampling=False,
min_num_negative_samples=1,
desired_negative_sampling_ratio=3,
use_keras=False,
predict_mask=False,
use_static_shapes=False,
nms_max_size_per_class=5):
is_training = False
num_classes = 1
mock_anchor_generator = MockAnchorGenerator2x2()
if use_keras:
mock_box_predictor = test_utils.MockKerasBoxPredictor(
is_training,
num_classes,
add_background_class=add_background_class,
predict_mask=predict_mask)
else:
mock_box_predictor = test_utils.MockBoxPredictor(
is_training,
num_classes,
add_background_class=add_background_class,
predict_mask=predict_mask)
mock_box_coder = test_utils.MockBoxCoder()
if use_keras:
fake_feature_extractor = FakeSSDKerasFeatureExtractor()
else:
fake_feature_extractor = FakeSSDFeatureExtractor()
mock_matcher = test_utils.MockMatcher()
region_similarity_calculator = sim_calc.IouSimilarity()
encode_background_as_zeros = False
def image_resizer_fn(image):
return [tf.identity(image), tf.shape(image)]
classification_loss = losses.WeightedSigmoidClassificationLoss()
localization_loss = losses.WeightedSmoothL1LocalizationLoss()
non_max_suppression_fn = functools.partial(
post_processing.batch_multiclass_non_max_suppression,
score_thresh=-20.0,
iou_thresh=1.0,
max_size_per_class=nms_max_size_per_class,
max_total_size=nms_max_size_per_class,
use_static_shapes=use_static_shapes)
classification_loss_weight = 1.0
localization_loss_weight = 1.0
negative_class_weight = 1.0
normalize_loss_by_num_matches = False
hard_example_miner = None
if apply_hard_mining:
# This hard example miner is expected to be a no-op.
hard_example_miner = losses.HardExampleMiner(
num_hard_examples=None, iou_threshold=1.0)
random_example_sampler = None
if random_example_sampling:
random_example_sampler = sampler.BalancedPositiveNegativeSampler(
positive_fraction=0.5)
target_assigner_instance = target_assigner.TargetAssigner(
region_similarity_calculator,
mock_matcher,
mock_box_coder,
negative_class_weight=negative_class_weight,
weight_regression_loss_by_score=weight_regression_loss_by_score)
expected_classification_loss_under_sampling = None
if use_expected_classification_loss_under_sampling:
expected_classification_loss_under_sampling = functools.partial(
ops.expected_classification_loss_under_sampling,
min_num_negative_samples=min_num_negative_samples,
desired_negative_sampling_ratio=desired_negative_sampling_ratio
)
code_size = 4
model = model_fn(
is_training=is_training,
anchor_generator=mock_anchor_generator,
box_predictor=mock_box_predictor,
box_coder=mock_box_coder,
feature_extractor=fake_feature_extractor,
encode_background_as_zeros=encode_background_as_zeros,
image_resizer_fn=image_resizer_fn,
non_max_suppression_fn=non_max_suppression_fn,
score_conversion_fn=tf.identity,
classification_loss=classification_loss,
localization_loss=localization_loss,
classification_loss_weight=classification_loss_weight,
localization_loss_weight=localization_loss_weight,
normalize_loss_by_num_matches=normalize_loss_by_num_matches,
hard_example_miner=hard_example_miner,
target_assigner_instance=target_assigner_instance,
add_summaries=False,
normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize,
freeze_batchnorm=False,
inplace_batchnorm_update=False,
add_background_class=add_background_class,
random_example_sampler=random_example_sampler,
expected_classification_loss_under_sampling=
expected_classification_loss_under_sampling)
return model, num_classes, mock_anchor_generator.num_anchors(
), code_size
