def test_mobilenet_creation(self, model_id, filter_size_scale):
"""Test creation of Mobilenet models."""
network = backbones.MobileNet(model_id=model_id,
filter_size_scale=filter_size_scale,
norm_momentum=0.99,
norm_epsilon=1e-5)
backbone_config = backbones_cfg.Backbone(
type='mobilenet',
mobilenet=backbones_cfg.MobileNet(
model_id=model_id, filter_size_scale=filter_size_scale))
norm_activation_config = common_cfg.NormActivation(norm_momentum=0.99,
norm_epsilon=1e-5,
use_sync_bn=False)
factory_network = factory.build_backbone(
input_specs=tf.keras.layers.InputSpec(shape=[None, None, None, 3]),
backbone_config=backbone_config,
norm_activation_config=norm_activation_config)
network_config = network.get_config()
factory_network_config = factory_network.get_config()
self.assertEqual(network_config, factory_network_config)
def test_spinenet_creation(self, model_id):
"""Test creation of SpineNet models."""
input_size = 128
min_level = 3
max_level = 7
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size, input_size, 3])
network = backbones.SpineNet(input_specs=input_specs,
min_level=min_level,
max_level=max_level,
norm_momentum=0.99,
norm_epsilon=1e-5)
backbone_config = backbones_cfg.Backbone(
type='spinenet',
spinenet=backbones_cfg.SpineNet(model_id=model_id))
norm_activation_config = common_cfg.NormActivation(norm_momentum=0.99,
norm_epsilon=1e-5,
use_sync_bn=False)
factory_network = factory.build_backbone(
input_specs=tf.keras.layers.InputSpec(
shape=[None, input_size, input_size, 3]),
backbone_config=backbone_config,
norm_activation_config=norm_activation_config)
network_config = network.get_config()
factory_network_config = factory_network.get_config()
self.assertEqual(network_config, factory_network_config)
def test_efficientnet_creation(self, model_id, se_ratio):
"""Test creation of EfficientNet models."""
network = backbones.EfficientNet(model_id=model_id,
se_ratio=se_ratio,
norm_momentum=0.99,
norm_epsilon=1e-5)
backbone_config = backbones_cfg.Backbone(
type='efficientnet',
efficientnet=backbones_cfg.EfficientNet(model_id=model_id,
se_ratio=se_ratio))
norm_activation_config = common_cfg.NormActivation(norm_momentum=0.99,
norm_epsilon=1e-5,
use_sync_bn=False)
factory_network = factory.build_backbone(
input_specs=tf.keras.layers.InputSpec(shape=[None, None, None, 3]),
backbone_config=backbone_config,
norm_activation_config=norm_activation_config)
network_config = network.get_config()
factory_network_config = factory_network.get_config()
self.assertEqual(network_config, factory_network_config)
def _create_centernet_model(model_id: int = 52,
num_hourglasses: int = 2
) -> centernet_model.CenterNetModel:
"""Create centernet model to load TF1 weights."""
task_config = centernet.CenterNetTask(
model=centernet.CenterNetModel(
backbone=backbones.Backbone(
type="hourglass",
hourglass=backbones.Hourglass(
model_id=model_id, num_hourglasses=num_hourglasses))))
model_config = task_config.model
backbone = factory.build_backbone(
input_specs=tf.keras.layers.InputSpec(shape=[1, 512, 512, 3]),
backbone_config=model_config.backbone,
norm_activation_config=model_config.norm_activation)
task_outputs = task_config.get_output_length_dict()
head = centernet_head.CenterNetHead(
input_specs=backbone.output_specs,
task_outputs=task_outputs,
input_levels=model_config.head.input_levels)
detect_generator_obj = detection_generator.CenterNetDetectionGenerator()
model = centernet_model.CenterNetModel(
backbone=backbone, head=head, detection_generator=detect_generator_obj)
logging.info("Successfully created centernet model.")
return model
def build_segmentation_model_3d(
input_specs: tf.keras.layers.InputSpec,
model_config: hyperparams.Config,
l2_regularizer: tf.keras.regularizers.Regularizer = None) -> tf.keras.Model: # pytype: disable=annotation-type-mismatch # typed-keras
"""Builds Segmentation model."""
norm_activation_config = model_config.norm_activation
backbone = backbone_factory.build_backbone(
input_specs=input_specs,
backbone_config=model_config.backbone,
norm_activation_config=norm_activation_config,
l2_regularizer=l2_regularizer)
decoder = decoder_factory.build_decoder(
input_specs=backbone.output_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
head_config = model_config.head
head = segmentation_heads_3d.SegmentationHead3D(
num_classes=model_config.num_classes,
level=head_config.level,
num_convs=head_config.num_convs,
num_filters=head_config.num_filters,
upsample_factor=head_config.upsample_factor,
activation=norm_activation_config.activation,
use_sync_bn=norm_activation_config.use_sync_bn,
norm_momentum=norm_activation_config.norm_momentum,
norm_epsilon=norm_activation_config.norm_epsilon,
use_batch_normalization=head_config.use_batch_normalization,
kernel_regularizer=l2_regularizer,
output_logits=head_config.output_logits)
model = segmentation_model.SegmentationModel(backbone, decoder, head)
return model
def build_yolo(input_specs, model_config, l2_regularization):
"""Builds yolo model."""
backbone = model_config.backbone.get()
anchor_dict, _ = model_config.anchor_boxes.get(backbone.min_level,
backbone.max_level)
backbone = backbone_factory.build_backbone(input_specs,
model_config.backbone,
model_config.norm_activation,
l2_regularization)
decoder = decoder_factory.build_decoder(backbone.output_specs,
model_config, l2_regularization)
head = build_yolo_head(decoder.output_specs, model_config,
l2_regularization)
detection_generator_obj = build_yolo_detection_generator(
model_config, anchor_dict)
model = yolo_model.Yolo(backbone=backbone,
decoder=decoder,
head=head,
detection_generator=detection_generator_obj)
model.build(input_specs.shape)
model.summary(print_fn=logging.info)
losses = detection_generator_obj.get_losses()
return model, losses
def test_resnet_creation(self, model_id):
"""Test creation of ResNet models."""
network = backbones.ResNet(model_id=model_id,
se_ratio=0.0,
norm_momentum=0.99,
norm_epsilon=1e-5)
backbone_config = backbones_cfg.Backbone(type='resnet',
resnet=backbones_cfg.ResNet(
model_id=model_id,
se_ratio=0.0))
norm_activation_config = common_cfg.NormActivation(norm_momentum=0.99,
norm_epsilon=1e-5,
use_sync_bn=False)
model_config = retinanet_cfg.RetinaNet(
backbone=backbone_config, norm_activation=norm_activation_config)
factory_network = factory.build_backbone(
input_specs=tf.keras.layers.InputSpec(shape=[None, None, None, 3]),
model_config=model_config)
network_config = network.get_config()
factory_network_config = factory_network.get_config()
self.assertEqual(network_config, factory_network_config)
def build_model(self):
"""get an instance of CenterNet."""
model_config = self.task_config.model
input_specs = tf.keras.layers.InputSpec(shape=[None] +
model_config.input_size)
l2_weight_decay = self.task_config.weight_decay
# Divide weight decay by 2.0 to match the implementation of tf.nn.l2_loss.
# (https://www.tensorflow.org/api_docs/python/tf/keras/regularizers/l2)
# (https://www.tensorflow.org/api_docs/python/tf/nn/l2_loss)
l2_regularizer = (tf.keras.regularizers.l2(l2_weight_decay / 2.0)
if l2_weight_decay else None)
backbone = factory.build_backbone(
input_specs=input_specs,
backbone_config=model_config.backbone,
norm_activation_config=model_config.norm_activation,
l2_regularizer=l2_regularizer)
task_outputs = self.task_config.get_output_length_dict()
head_config = model_config.head
head = centernet_head.CenterNetHead(
input_specs=backbone.output_specs,
task_outputs=task_outputs,
input_levels=head_config.input_levels,
heatmap_bias=head_config.heatmap_bias)
# output_specs is a dict
backbone_output_spec = backbone.output_specs[
head_config.input_levels[-1]]
if len(backbone_output_spec) == 4:
bb_output_height = backbone_output_spec[1]
elif len(backbone_output_spec) == 3:
bb_output_height = backbone_output_spec[0]
else:
raise ValueError
self._net_down_scale = int(model_config.input_size[0] /
bb_output_height)
dg_config = model_config.detection_generator
detect_generator_obj = detection_generator.CenterNetDetectionGenerator(
max_detections=dg_config.max_detections,
peak_error=dg_config.peak_error,
peak_extract_kernel_size=dg_config.peak_extract_kernel_size,
class_offset=dg_config.class_offset,
net_down_scale=self._net_down_scale,
input_image_dims=model_config.input_size[0],
use_nms=dg_config.use_nms,
nms_pre_thresh=dg_config.nms_pre_thresh,
nms_thresh=dg_config.nms_thresh)
model = centernet_model.CenterNetModel(
backbone=backbone,
head=head,
detection_generator=detect_generator_obj)
return model
def build_yolo(input_specs, model_config, l2_regularization, masks, xy_scales,
path_scales):
print(model_config.as_dict())
print(input_specs)
print(l2_regularization)
backbone = factory.build_backbone(input_specs, model_config,
l2_regularization)
decoder = build_yolo_decoder(backbone.output_specs, model_config,
l2_regularization)
head = build_yolo_head(backbone.output_specs, model_config,
l2_regularization)
filter = build_yolo_filter(model_config, head, masks, xy_scales,
path_scales)
model = Yolo(backbone=backbone, decoder=decoder, head=head, filter=filter)
model.build(input_specs.shape)
losses = filter.losses
return model, losses
