def _create_model(
self,
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):
return super(SsdMetaArchTest, self)._create_model(
model_fn=ssd_meta_arch.SSDMetaArch,
apply_hard_mining=apply_hard_mining,
normalize_loc_loss_by_codesize=normalize_loc_loss_by_codesize,
add_background_class=add_background_class,
random_example_sampling=random_example_sampling,
expected_loss_weights=expected_loss_weights,
min_num_negative_samples=min_num_negative_samples,
desired_negative_sampling_ratio=desired_negative_sampling_ratio,
use_keras=use_keras,
predict_mask=predict_mask,
use_static_shapes=use_static_shapes,
nms_max_size_per_class=nms_max_size_per_class)
def testGetImageResizerConfig(self):
"""Tests that number of classes can be retrieved."""
model_config = model_pb2.DetectionModel()
model_config.faster_rcnn.image_resizer.fixed_shape_resizer.height = 100
model_config.faster_rcnn.image_resizer.fixed_shape_resizer.width = 300
image_resizer_config = config_util.get_image_resizer_config(
model_config)
self.assertEqual(image_resizer_config.fixed_shape_resizer.height, 100)
self.assertEqual(image_resizer_config.fixed_shape_resizer.width, 300)
def get_configs_from_multiple_files(model_config_path="",
train_config_path="",
train_input_config_path="",
eval_config_path="",
eval_input_config_path="",
graph_rewriter_config_path=""):
"""Reads training configuration from multiple config files.
Args:
model_config_path: Path to model_pb2.DetectionModel.
train_config_path: Path to train_pb2.TrainConfig.
train_input_config_path: Path to input_reader_pb2.InputReader.
eval_config_path: Path to eval_pb2.EvalConfig.
eval_input_config_path: Path to input_reader_pb2.InputReader.
graph_rewriter_config_path: Path to graph_rewriter_pb2.GraphRewriter.
Returns:
Dictionary of configuration objects. Keys are `model`, `train_config`,
`train_input_config`, `eval_config`, `eval_input_config`. Key/Values are
returned only for valid (non-empty) strings.
"""
configs = {}
if model_config_path:
model_config = model_pb2.DetectionModel()
with tf.gfile.GFile(model_config_path, "r") as f:
text_format.Merge(f.read(), model_config)
configs["model"] = model_config
if train_config_path:
train_config = train_pb2.TrainConfig()
with tf.gfile.GFile(train_config_path, "r") as f:
text_format.Merge(f.read(), train_config)
configs["train_config"] = train_config
if train_input_config_path:
train_input_config = input_reader_pb2.InputReader()
with tf.gfile.GFile(train_input_config_path, "r") as f:
text_format.Merge(f.read(), train_input_config)
configs["train_input_config"] = train_input_config
if eval_config_path:
eval_config = eval_pb2.EvalConfig()
with tf.gfile.GFile(eval_config_path, "r") as f:
text_format.Merge(f.read(), eval_config)
configs["eval_config"] = eval_config
if eval_input_config_path:
eval_input_config = input_reader_pb2.InputReader()
with tf.gfile.GFile(eval_input_config_path, "r") as f:
text_format.Merge(f.read(), eval_input_config)
configs["eval_input_configs"] = [eval_input_config]
if graph_rewriter_config_path:
configs["graph_rewriter_config"] = get_graph_rewriter_config_from_file(
graph_rewriter_config_path)
return configs
def test_create_faster_rcnn_model_from_config_with_example_miner(self):
model_text_proto = """
faster_rcnn {
num_classes: 3
feature_extractor {
type: 'faster_rcnn_inception_resnet_v2'
}
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 600
max_dimension: 1024
}
}
first_stage_anchor_generator {
grid_anchor_generator {
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 2.0]
height_stride: 16
width_stride: 16
}
}
first_stage_box_predictor_conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
second_stage_box_predictor {
mask_rcnn_box_predictor {
fc_hyperparams {
op: FC
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
hard_example_miner {
num_hard_examples: 10
iou_threshold: 0.99
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model = model_builder.build(model_proto, is_training=True)
self.assertIsNotNone(model._hard_example_miner)
def test_get_configs_from_multiple_files(self):
"""Tests that proto configs can be read from multiple files."""
temp_dir = self.get_temp_dir()
# Write model config file.
model_config_path = os.path.join(temp_dir, "model.config")
model = model_pb2.DetectionModel()
model.faster_rcnn.num_classes = 10
_write_config(model, model_config_path)
# Write train config file.
train_config_path = os.path.join(temp_dir, "train.config")
train_config = train_config = train_pb2.TrainConfig()
train_config.batch_size = 32
_write_config(train_config, train_config_path)
# Write train input config file.
train_input_config_path = os.path.join(temp_dir, "train_input.config")
train_input_config = input_reader_pb2.InputReader()
train_input_config.label_map_path = "path/to/label_map"
_write_config(train_input_config, train_input_config_path)
# Write eval config file.
eval_config_path = os.path.join(temp_dir, "eval.config")
eval_config = eval_pb2.EvalConfig()
eval_config.num_examples = 20
_write_config(eval_config, eval_config_path)
# Write eval input config file.
eval_input_config_path = os.path.join(temp_dir, "eval_input.config")
eval_input_config = input_reader_pb2.InputReader()
eval_input_config.label_map_path = "path/to/another/label_map"
_write_config(eval_input_config, eval_input_config_path)
configs = config_util.get_configs_from_multiple_files(
model_config_path=model_config_path,
train_config_path=train_config_path,
train_input_config_path=train_input_config_path,
eval_config_path=eval_config_path,
eval_input_config_path=eval_input_config_path)
self.assertProtoEquals(model, configs["model"])
self.assertProtoEquals(train_config, configs["train_config"])
self.assertProtoEquals(train_input_config,
configs["train_input_config"])
self.assertProtoEquals(eval_config, configs["eval_config"])
self.assertProtoEquals(eval_input_config,
configs["eval_input_configs"][0])
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):
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)
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=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_loss_weights_fn=expected_loss_weights_fn,
**kwargs)
return model, num_classes, mock_anchor_generator.num_anchors(), code_size
def test_create_embedded_ssd_mobilenet_v1_model_from_config(self):
model_text_proto = """
ssd {
feature_extractor {
type: 'embedded_ssd_mobilenet_v1'
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
box_coder {
faster_rcnn_box_coder {
}
}
matcher {
argmax_matcher {
}
}
similarity_calculator {
iou_similarity {
}
}
anchor_generator {
ssd_anchor_generator {
aspect_ratios: 1.0
}
}
image_resizer {
fixed_shape_resizer {
height: 256
width: 256
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
loss {
classification_loss {
weighted_softmax {
}
}
localization_loss {
weighted_smooth_l1 {
}
}
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model = self.create_model(model_proto)
self.assertIsInstance(model, ssd_meta_arch.SSDMetaArch)
self.assertIsInstance(model._feature_extractor,
EmbeddedSSDMobileNetV1FeatureExtractor)
def test_create_ssd_mobilenet_v2_fpnlite_model_from_config(self):
model_text_proto = """
ssd {
freeze_batchnorm: true
inplace_batchnorm_update: true
feature_extractor {
type: 'ssd_mobilenet_v2_fpn'
use_depthwise: true
fpn {
min_level: 3
max_level: 7
additional_layer_depth: 128
}
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
box_coder {
faster_rcnn_box_coder {
}
}
matcher {
argmax_matcher {
}
}
similarity_calculator {
iou_similarity {
}
}
anchor_generator {
ssd_anchor_generator {
aspect_ratios: 1.0
}
}
image_resizer {
fixed_shape_resizer {
height: 320
width: 320
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
normalize_loc_loss_by_codesize: true
loss {
classification_loss {
weighted_softmax {
}
}
localization_loss {
weighted_smooth_l1 {
}
}
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model = self.create_model(model_proto)
self.assertIsInstance(model, ssd_meta_arch.SSDMetaArch)
self.assertIsInstance(model._feature_extractor,
SSDMobileNetV2FpnFeatureExtractor)
self.assertTrue(model._normalize_loc_loss_by_codesize)
self.assertTrue(model._freeze_batchnorm)
self.assertTrue(model._inplace_batchnorm_update)
def test_create_ssd_inception_v2_model_from_config(self):
model_text_proto = """
ssd {
feature_extractor {
type: 'ssd_inception_v2'
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
override_base_feature_extractor_hyperparams: true
}
box_coder {
faster_rcnn_box_coder {
}
}
matcher {
argmax_matcher {
}
}
similarity_calculator {
iou_similarity {
}
}
anchor_generator {
ssd_anchor_generator {
aspect_ratios: 1.0
}
}
image_resizer {
fixed_shape_resizer {
height: 320
width: 320
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
loss {
classification_loss {
weighted_softmax {
}
}
localization_loss {
weighted_smooth_l1 {
}
}
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model = self.create_model(model_proto)
self.assertIsInstance(model, ssd_meta_arch.SSDMetaArch)
self.assertIsInstance(model._feature_extractor,
SSDInceptionV2FeatureExtractor)
self.assertIsNone(model._expected_loss_weights_fn)
def test_create_ssd_resnet_v1_ppn_model_from_config(self):
model_text_proto = """
ssd {
feature_extractor {
type: 'ssd_resnet_v1_50_ppn'
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
box_coder {
mean_stddev_box_coder {
}
}
matcher {
bipartite_matcher {
}
}
similarity_calculator {
iou_similarity {
}
}
anchor_generator {
ssd_anchor_generator {
aspect_ratios: 1.0
}
}
image_resizer {
fixed_shape_resizer {
height: 320
width: 320
}
}
box_predictor {
weight_shared_convolutional_box_predictor {
depth: 1024
class_prediction_bias_init: -4.6
conv_hyperparams {
activation: RELU_6,
regularizer {
l2_regularizer {
weight: 0.0004
}
}
initializer {
variance_scaling_initializer {
}
}
}
num_layers_before_predictor: 2
kernel_size: 1
}
}
loss {
classification_loss {
weighted_softmax {
}
}
localization_loss {
weighted_l2 {
}
}
classification_weight: 1.0
localization_weight: 1.0
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
for extractor_type, extractor_class in SSD_RESNET_V1_PPN_FEAT_MAPS.items(
):
model_proto.ssd.feature_extractor.type = extractor_type
model = model_builder.build(model_proto, is_training=True)
self.assertIsInstance(model, ssd_meta_arch.SSDMetaArch)
self.assertIsInstance(model._feature_extractor, extractor_class)
def test_create_ssd_resnet_v1_fpn_model_from_config(self):
model_text_proto = """
ssd {
feature_extractor {
type: 'ssd_resnet50_v1_fpn'
fpn {
min_level: 3
max_level: 7
}
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
box_coder {
faster_rcnn_box_coder {
}
}
matcher {
argmax_matcher {
}
}
similarity_calculator {
iou_similarity {
}
}
encode_background_as_zeros: true
anchor_generator {
multiscale_anchor_generator {
aspect_ratios: [1.0, 2.0, 0.5]
scales_per_octave: 2
}
}
image_resizer {
fixed_shape_resizer {
height: 320
width: 320
}
}
box_predictor {
weight_shared_convolutional_box_predictor {
depth: 32
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
random_normal_initializer {
}
}
}
num_layers_before_predictor: 1
}
}
normalize_loss_by_num_matches: true
normalize_loc_loss_by_codesize: true
loss {
classification_loss {
weighted_sigmoid_focal {
alpha: 0.25
gamma: 2.0
}
}
localization_loss {
weighted_smooth_l1 {
delta: 0.1
}
}
classification_weight: 1.0
localization_weight: 1.0
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
for extractor_type, extractor_class in SSD_RESNET_V1_FPN_FEAT_MAPS.items(
):
model_proto.ssd.feature_extractor.type = extractor_type
model = model_builder.build(model_proto, is_training=True)
self.assertIsInstance(model, ssd_meta_arch.SSDMetaArch)
self.assertIsInstance(model._feature_extractor, extractor_class)
def test_create_rfcn_resnet_v1_model_from_config(self):
model_text_proto = """
faster_rcnn {
num_classes: 3
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 600
max_dimension: 1024
}
}
feature_extractor {
type: 'faster_rcnn_resnet101'
}
first_stage_anchor_generator {
grid_anchor_generator {
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 2.0]
height_stride: 16
width_stride: 16
}
}
first_stage_box_predictor_conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
initial_crop_size: 14
maxpool_kernel_size: 2
maxpool_stride: 2
second_stage_box_predictor {
rfcn_box_predictor {
conv_hyperparams {
op: CONV
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
second_stage_post_processing {
batch_non_max_suppression {
score_threshold: 0.01
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 300
}
score_converter: SOFTMAX
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
for extractor_type, extractor_class in FRCNN_RESNET_FEAT_MAPS.items():
model_proto.faster_rcnn.feature_extractor.type = extractor_type
model = model_builder.build(model_proto, is_training=True)
self.assertIsInstance(model, rfcn_meta_arch.RFCNMetaArch)
self.assertIsInstance(model._feature_extractor, extractor_class)
def test_create_faster_rcnn_inception_v2_model_from_config(self):
model_text_proto = """
faster_rcnn {
num_classes: 3
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 600
max_dimension: 1024
}
}
feature_extractor {
type: 'faster_rcnn_inception_v2'
}
first_stage_anchor_generator {
grid_anchor_generator {
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 2.0]
height_stride: 16
width_stride: 16
}
}
first_stage_box_predictor_conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
initial_crop_size: 14
maxpool_kernel_size: 2
maxpool_stride: 2
second_stage_box_predictor {
mask_rcnn_box_predictor {
fc_hyperparams {
op: FC
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
}
}
second_stage_post_processing {
batch_non_max_suppression {
score_threshold: 0.01
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 300
}
score_converter: SOFTMAX
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model = model_builder.build(model_proto, is_training=True)
self.assertIsInstance(model, faster_rcnn_meta_arch.FasterRCNNMetaArch)
self.assertIsInstance(
model._feature_extractor,
frcnn_inc_v2.FasterRCNNInceptionV2FeatureExtractor)
def test_create_faster_rcnn_resnet101_with_mask_prediction_enabled(
self, use_matmul_crop_and_resize):
model_text_proto = """
faster_rcnn {
num_classes: 3
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 600
max_dimension: 1024
}
}
feature_extractor {
type: 'faster_rcnn_resnet101'
}
first_stage_anchor_generator {
grid_anchor_generator {
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 2.0]
height_stride: 16
width_stride: 16
}
}
first_stage_box_predictor_conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
initial_crop_size: 14
maxpool_kernel_size: 2
maxpool_stride: 2
second_stage_box_predictor {
mask_rcnn_box_predictor {
fc_hyperparams {
op: FC
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
conv_hyperparams {
regularizer {
l2_regularizer {
}
}
initializer {
truncated_normal_initializer {
}
}
}
predict_instance_masks: true
}
}
second_stage_mask_prediction_loss_weight: 3.0
second_stage_post_processing {
batch_non_max_suppression {
score_threshold: 0.01
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 300
}
score_converter: SOFTMAX
}
}"""
model_proto = model_pb2.DetectionModel()
text_format.Merge(model_text_proto, model_proto)
model_proto.faster_rcnn.use_matmul_crop_and_resize = (
use_matmul_crop_and_resize)
model = model_builder.build(model_proto, is_training=True)
self.assertAlmostEqual(model._second_stage_mask_loss_weight, 3.0)
