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 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_logits_softmax_classification_loss(self): losses_text_proto = """ classification_loss { weighted_logits_softmax { } } 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.WeightedSoftmaxClassificationAgainstLogitsLoss))
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_smooth_l1_localization_loss_default_delta(self): losses_text_proto = """ localization_loss { weighted_smooth_l1 { } } 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.WeightedSmoothL1LocalizationLoss)) self.assertAlmostEqual(localization_loss._delta, 1.0)
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 = """ localization_loss { weighted_smooth_l1 { } } 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.WeightedSmoothL1LocalizationLoss)) predictions = tf.constant([[[0.0, 0.0, 1.0, 1.0], [0.0, 0.0, 1.0, 1.0]]]) targets = tf.constant([[[0.0, 0.0, 1.0, 1.0], [0.0, 0.0, 1.0, 1.0]]]) weights = tf.constant([[1.0, 1.0]]) loss = localization_loss(predictions, targets, weights=weights) self.assertEqual(loss.shape, [1, 2])
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, 3])
def _build_ssd_model(ssd_config, is_training, add_summaries, add_background_class=True): """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. add_summaries: Whether to add tf summaries in the model. add_background_class: Whether to add an implicit background class to one-hot encodings of groundtruth labels. Set to false if using groundtruth labels with an explicit background class or using multiclass scores instead of truth in the case of distillation. 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( feature_extractor_config=ssd_config.feature_extractor, is_training=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) encode_background_as_zeros = ssd_config.encode_background_as_zeros negative_class_weight = ssd_config.negative_class_weight 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, random_example_sampler) = losses_builder.build(ssd_config.loss) normalize_loss_by_num_matches = ssd_config.normalize_loss_by_num_matches normalize_loc_loss_by_codesize = ssd_config.normalize_loc_loss_by_codesize return ssd_meta_arch.SSDMetaArch( is_training, anchor_generator, ssd_box_predictor, box_coder, feature_extractor, matcher, region_similarity_calculator, encode_background_as_zeros, negative_class_weight, 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, add_summaries=add_summaries, normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize, freeze_batchnorm=ssd_config.freeze_batchnorm, inplace_batchnorm_update=ssd_config.inplace_batchnorm_update, add_background_class=add_background_class, random_example_sampler=random_example_sampler)