def test_builder(self, backbone_type, input_size, has_att_heads):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
if has_att_heads:
attribute_heads_config = [
retinanet_cfg.AttributeHead(name='att1'),
retinanet_cfg.AttributeHead(name='att2',
type='classification',
size=2),
]
else:
attribute_heads_config = None
model_config = retinanet_cfg.RetinaNet(
num_classes=num_classes,
backbone=backbones.Backbone(type=backbone_type),
head=retinanet_cfg.RetinaNetHead(
attribute_heads=attribute_heads_config))
l2_regularizer = tf.keras.regularizers.l2(5e-5)
_ = factory.build_retinanet(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
if has_att_heads:
self.assertEqual(model_config.head.attribute_heads[0].as_dict(),
dict(name='att1', type='regression', size=1))
self.assertEqual(model_config.head.attribute_heads[1].as_dict(),
dict(name='att2', type='classification', size=2))
def test_builder(self, input_size, backbone_type, level, low_level,
decoder_type, shared_decoder, generate_panoptic_masks):
num_classes = 10
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
model_config = panoptic_deeplab_cfg.PanopticDeeplab(
num_classes=num_classes,
input_size=input_size,
backbone=backbones.Backbone(type=backbone_type),
decoder=decoders.Decoder(type=decoder_type),
semantic_head=panoptic_deeplab_cfg.SemanticHead(
level=level,
num_convs=1,
kernel_size=5,
prediction_kernel_size=1,
low_level=low_level),
instance_head=panoptic_deeplab_cfg.InstanceHead(
level=level,
num_convs=1,
kernel_size=5,
prediction_kernel_size=1,
low_level=low_level),
shared_decoder=shared_decoder,
generate_panoptic_masks=generate_panoptic_masks)
l2_regularizer = tf.keras.regularizers.l2(5e-5)
_ = factory.build_panoptic_deeplab(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
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_validation_step(self):
config = detr_cfg.DetrTask(
model=detr_cfg.Detr(
input_size=[1333, 1333, 3],
num_encoder_layers=1,
num_decoder_layers=1,
backbone=backbones.Backbone(
type='resnet',
resnet=backbones.ResNet(model_id=10, bn_trainable=False))
),
losses=detr_cfg.Losses(class_offset=1),
validation_data=detr_cfg.DataConfig(
tfds_name='coco/2017',
tfds_split='validation',
is_training=False,
global_batch_size=2,
))
with tfds.testing.mock_data(as_dataset_fn=_as_dataset):
task = detection.DetectionTask(config)
model = task.build_model()
metrics = task.build_metrics(training=False)
dataset = task.build_inputs(config.validation_data)
iterator = iter(dataset)
logs = task.validation_step(next(iterator), model, metrics)
state = task.aggregate_logs(step_outputs=logs)
task.reduce_aggregated_logs(state)
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 test_deeplabv3_builder(self, backbone_type, input_size, weight_decay):
num_classes = 21
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
model_config = semantic_segmentation_cfg.SemanticSegmentationModel(
num_classes=num_classes,
backbone=backbones.Backbone(type=backbone_type,
mobilenet=backbones.MobileNet(
model_id='MobileNetV2',
output_stride=16)),
decoder=decoders.Decoder(type='aspp',
aspp=decoders.ASPP(level=4,
num_filters=256,
dilation_rates=[],
spp_layer_version='v1',
output_tensor=True)),
head=semantic_segmentation_cfg.SegmentationHead(
level=4,
low_level=2,
num_convs=1,
upsample_factor=2,
use_depthwise_convolution=True))
l2_regularizer = (tf.keras.regularizers.l2(weight_decay)
if weight_decay else None)
model = factory.build_segmentation_model(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
quantization_config = common.Quantization()
_ = qat_factory.build_qat_segmentation_model(
model=model,
quantization=quantization_config,
input_specs=input_specs)
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_builder(self, backbone_type, input_size):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
model_config = maskrcnn_cfg.MaskRCNN(
num_classes=num_classes,
backbone=backbones.Backbone(type=backbone_type))
l2_regularizer = tf.keras.regularizers.l2(5e-5)
_ = factory.build_maskrcnn(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
def test_builder(self, backbone_type, input_size, weight_decay):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
model_config = classification_cfg.ImageClassificationModel(
num_classes=num_classes,
backbone=backbones.Backbone(type=backbone_type))
l2_regularizer = (tf.keras.regularizers.l2(weight_decay)
if weight_decay else None)
_ = factory.build_classification_model(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
class Detr(hyperparams.Config):
num_queries: int = 100
hidden_size: int = 256
num_classes: int = 91 # 0: background
num_encoder_layers: int = 6
num_decoder_layers: int = 6
input_size: List[int] = dataclasses.field(default_factory=list)
backbone: backbones.Backbone = backbones.Backbone(type='resnet',
resnet=backbones.ResNet(
model_id=50,
bn_trainable=False))
norm_activation: common.NormActivation = common.NormActivation()
class SimCLRModel(hyperparams.Config):
"""SimCLR model config."""
input_size: List[int] = dataclasses.field(default_factory=list)
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
projection_head: ProjectionHead = ProjectionHead(
proj_output_dim=128, num_proj_layers=3, ft_proj_idx=1)
supervised_head: SupervisedHead = SupervisedHead(num_classes=1001)
norm_activation: common.NormActivation = common.NormActivation(
norm_momentum=0.9, norm_epsilon=1e-5, use_sync_bn=False)
mode: str = simclr_model.PRETRAIN
backbone_trainable: bool = True
class SemanticSegmentationModel(hyperparams.Config):
"""Semantic segmentation model config."""
num_classes: int = 0
input_size: List[int] = dataclasses.field(default_factory=list)
min_level: int = 3
max_level: int = 6
head: SegmentationHead = SegmentationHead()
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
decoder: decoders.Decoder = decoders.Decoder(type='identity')
mask_scoring_head: Optional[MaskScoringHead] = None
norm_activation: common.NormActivation = common.NormActivation()
class ImageClassificationModel(hyperparams.Config):
"""The model config."""
num_classes: int = 0
input_size: List[int] = dataclasses.field(default_factory=list)
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
dropout_rate: float = 0.0
norm_activation: common.NormActivation = common.NormActivation(
use_sync_bn=False)
# Adds a BatchNormalization layer pre-GlobalAveragePooling in classification
add_head_batch_norm: bool = False
kernel_initializer: str = 'random_uniform'
class RetinaNet(hyperparams.Config):
num_classes: int = 0
input_size: List[int] = dataclasses.field(default_factory=list)
min_level: int = 3
max_level: int = 7
anchor: Anchor = Anchor()
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
decoder: decoders.Decoder = decoders.Decoder(type='fpn',
fpn=decoders.FPN())
head: RetinaNetHead = RetinaNetHead()
detection_generator: DetectionGenerator = DetectionGenerator()
norm_activation: common.NormActivation = common.NormActivation()
def test_builder(self, backbone_type, input_size,
segmentation_backbone_type, segmentation_decoder_type):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
segmentation_output_stride = 16
level = int(np.math.log2(segmentation_output_stride))
segmentation_model = semantic_segmentation.SemanticSegmentationModel(
num_classes=2,
backbone=backbones.Backbone(type=segmentation_backbone_type),
decoder=decoders.Decoder(type=segmentation_decoder_type),
head=semantic_segmentation.SegmentationHead(level=level))
model_config = panoptic_maskrcnn_cfg.PanopticMaskRCNN(
num_classes=num_classes,
segmentation_model=segmentation_model,
backbone=backbones.Backbone(type=backbone_type),
shared_backbone=segmentation_backbone_type is None,
shared_decoder=segmentation_decoder_type is None)
l2_regularizer = tf.keras.regularizers.l2(5e-5)
_ = factory.build_panoptic_maskrcnn(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
class PanopticDeeplab(hyperparams.Config):
"""Panoptic Deeplab model config."""
num_classes: int = 2
input_size: List[int] = dataclasses.field(default_factory=list)
min_level: int = 3
max_level: int = 6
norm_activation: common.NormActivation = common.NormActivation()
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
decoder: decoders.Decoder = decoders.Decoder(type='aspp')
semantic_head: SemanticHead = SemanticHead()
instance_head: InstanceHead = InstanceHead()
shared_decoder: bool = False
generate_panoptic_masks: bool = True
post_processor: PanopticDeeplabPostProcessor = PanopticDeeplabPostProcessor(
)
def retinanet_spinenet_mobile_coco() -> cfg.ExperimentConfig:
"""Generates a config for COCO OD RetinaNet for mobile with QAT."""
config = retinanet.retinanet_spinenet_mobile_coco()
task = RetinaNetTask.from_args(quantization=common.Quantization(),
**config.task.as_dict())
task.model.backbone = backbones.Backbone(
type='spinenet_mobile',
spinenet_mobile=backbones.SpineNetMobile(
model_id='49',
stochastic_depth_drop_rate=0.2,
min_level=3,
max_level=7,
use_keras_upsampling_2d=True))
config.task = task
return config
class SimCLRMTModelConfig(hyperparams.Config):
"""Model config for multi-task SimCLR model."""
input_size: List[int] = dataclasses.field(default_factory=list)
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
backbone_trainable: bool = True
projection_head: simclr_configs.ProjectionHead = simclr_configs.ProjectionHead(
proj_output_dim=128, num_proj_layers=3, ft_proj_idx=1)
norm_activation: common.NormActivation = common.NormActivation(
norm_momentum=0.9, norm_epsilon=1e-5, use_sync_bn=False)
heads: Tuple[SimCLRMTHeadConfig, ...] = ()
# L2 weight decay is used in the model, not in task.
# Note that this can not be used together with lars optimizer.
l2_weight_decay: float = 0.0
init_checkpoint: str = ''
# backbone_projection or backbone
init_checkpoint_modules: str = 'backbone_projection'
class AutosegEdgeTPUModelConfig(hyperparams.Config):
"""Autoseg-EdgeTPU segmentation model config."""
num_classes: int = 0
input_size: List[int] = dataclasses.field(default_factory=list)
backbone: backbones.Backbone = backbones.Backbone()
head: BiFPNHeadConfig = BiFPNHeadConfig()
model_params: Mapping[str, Any] = dataclasses.field(
default_factory=lambda: { # pylint: disable=g-long-lambda
'model_name': 'autoseg_edgetpu_backbone_s',
'checkpoint_format': 'tf_checkpoint',
'overrides': {
'batch_norm': 'tpu',
'rescale_input': False,
'backbone_only': True,
'resolution': 512
}
})
def test_train_step(self):
config = detr_cfg.DetrTask(
model=detr_cfg.Detr(
input_size=[1333, 1333, 3],
num_encoder_layers=1,
num_decoder_layers=1,
num_classes=81,
backbone=backbones.Backbone(
type='resnet',
resnet=backbones.ResNet(model_id=10, bn_trainable=False))
),
train_data=coco.COCODataConfig(
tfds_name='coco/2017',
tfds_split='validation',
is_training=True,
global_batch_size=2,
))
with tfds.testing.mock_data(as_dataset_fn=_as_dataset):
task = detection.DetectionTask(config)
model = task.build_model()
dataset = task.build_inputs(config.train_data)
iterator = iter(dataset)
opt_cfg = optimization.OptimizationConfig({
'optimizer': {
'type': 'detr_adamw',
'detr_adamw': {
'weight_decay_rate': 1e-4,
'global_clipnorm': 0.1,
}
},
'learning_rate': {
'type': 'stepwise',
'stepwise': {
'boundaries': [120000],
'values': [0.0001, 1.0e-05]
}
},
})
optimizer = detection.DetectionTask.create_optimizer(opt_cfg)
task.train_step(next(iterator), model, optimizer)
def test_builder(self, backbone_type, input_size, has_attribute_heads):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
if has_attribute_heads:
attribute_heads_config = [
retinanet_cfg.AttributeHead(name='att1'),
retinanet_cfg.AttributeHead(name='att2',
type='classification',
size=2),
]
else:
attribute_heads_config = None
model_config = retinanet_cfg.RetinaNet(
num_classes=num_classes,
backbone=backbones.Backbone(
type=backbone_type,
spinenet_mobile=backbones.SpineNetMobile(
model_id='49',
stochastic_depth_drop_rate=0.2,
min_level=3,
max_level=7,
use_keras_upsampling_2d=True)),
head=retinanet_cfg.RetinaNetHead(
attribute_heads=attribute_heads_config))
l2_regularizer = tf.keras.regularizers.l2(5e-5)
quantization_config = common.Quantization()
model = factory.build_retinanet(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
_ = qat_factory.build_qat_retinanet(model=model,
quantization=quantization_config,
model_config=model_config)
if has_attribute_heads:
self.assertEqual(model_config.head.attribute_heads[0].as_dict(),
dict(name='att1', type='regression', size=1))
self.assertEqual(model_config.head.attribute_heads[1].as_dict(),
dict(name='att2', type='classification', size=2))
class MaskRCNN(hyperparams.Config):
num_classes: int = 0
input_size: List[int] = dataclasses.field(default_factory=list)
min_level: int = 2
max_level: int = 6
anchor: Anchor = Anchor()
include_mask: bool = True
backbone: backbones.Backbone = backbones.Backbone(
type='resnet', resnet=backbones.ResNet())
decoder: decoders.Decoder = decoders.Decoder(type='fpn',
fpn=decoders.FPN())
rpn_head: RPNHead = RPNHead()
detection_head: DetectionHead = DetectionHead()
roi_generator: ROIGenerator = ROIGenerator()
roi_sampler: ROISampler = ROISampler()
roi_aligner: ROIAligner = ROIAligner()
detection_generator: DetectionGenerator = DetectionGenerator()
mask_head: Optional[MaskHead] = MaskHead()
mask_sampler: Optional[MaskSampler] = MaskSampler()
mask_roi_aligner: Optional[MaskROIAligner] = MaskROIAligner()
norm_activation: common.NormActivation = common.NormActivation(
norm_momentum=0.997, norm_epsilon=0.0001, use_sync_bn=True)
def image_classification_imagenet_mobilenet() -> cfg.ExperimentConfig:
"""Image classification on imagenet with mobilenet."""
train_batch_size = 4096
eval_batch_size = 4096
steps_per_epoch = IMAGENET_TRAIN_EXAMPLES // train_batch_size
config = cfg.ExperimentConfig(
task=ImageClassificationTask(
model=ImageClassificationModel(
num_classes=1001,
dropout_rate=0.2,
input_size=[224, 224, 3],
backbone=backbones.Backbone(type='mobilenet',
mobilenet=backbones.MobileNet(
model_id='MobileNetV2',
filter_size_scale=1.0)),
norm_activation=common.NormActivation(norm_momentum=0.997,
norm_epsilon=1e-3,
use_sync_bn=False)),
losses=Losses(l2_weight_decay=1e-5, label_smoothing=0.1),
train_data=DataConfig(input_path=os.path.join(
IMAGENET_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size),
validation_data=DataConfig(input_path=os.path.join(
IMAGENET_INPUT_PATH_BASE, 'valid*'),
is_training=False,
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
train_steps=500 * steps_per_epoch,
validation_steps=IMAGENET_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'rmsprop',
'rmsprop': {
'rho': 0.9,
'momentum': 0.9,
'epsilon': 0.002,
}
},
'learning_rate': {
'type': 'exponential',
'exponential': {
'initial_learning_rate':
0.008 * (train_batch_size // 128),
'decay_steps': int(2.5 * steps_per_epoch),
'decay_rate': 0.98,
'staircase': True
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 5 * steps_per_epoch,
'warmup_learning_rate': 0
}
},
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
])
return config
def simclr_finetuning_imagenet() -> cfg.ExperimentConfig:
"""Image classification general."""
train_batch_size = 1024
eval_batch_size = 1024
steps_per_epoch = IMAGENET_TRAIN_EXAMPLES // train_batch_size
pretrain_model_base = ''
return cfg.ExperimentConfig(
task=SimCLRFinetuneTask(
model=SimCLRModel(
mode=simclr_model.FINETUNE,
backbone_trainable=True,
input_size=[224, 224, 3],
backbone=backbones.Backbone(
type='resnet', resnet=backbones.ResNet(model_id=50)),
projection_head=ProjectionHead(
proj_output_dim=128, num_proj_layers=3, ft_proj_idx=1),
supervised_head=SupervisedHead(num_classes=1001, zero_init=True),
norm_activation=common.NormActivation(
norm_momentum=0.9, norm_epsilon=1e-5, use_sync_bn=False)),
loss=ClassificationLosses(),
evaluation=Evaluation(),
train_data=DataConfig(
parser=Parser(mode=simclr_model.FINETUNE),
input_path=os.path.join(IMAGENET_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size),
validation_data=DataConfig(
parser=Parser(mode=simclr_model.FINETUNE),
input_path=os.path.join(IMAGENET_INPUT_PATH_BASE, 'valid*'),
is_training=False,
global_batch_size=eval_batch_size),
init_checkpoint=pretrain_model_base,
# all, backbone_projection or backbone
init_checkpoint_modules='backbone_projection'),
trainer=cfg.TrainerConfig(
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
train_steps=60 * steps_per_epoch,
validation_steps=IMAGENET_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'lars',
'lars': {
'momentum':
0.9,
'weight_decay_rate':
0.0,
'exclude_from_weight_decay': [
'batch_normalization', 'bias'
]
}
},
'learning_rate': {
'type': 'cosine',
'cosine': {
# 0.01 × BatchSize / 512
'initial_learning_rate': 0.01 * train_batch_size / 512,
'decay_steps': 60 * steps_per_epoch
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
])
def simclr_pretraining_imagenet() -> cfg.ExperimentConfig:
"""Image classification general."""
train_batch_size = 4096
eval_batch_size = 4096
steps_per_epoch = IMAGENET_TRAIN_EXAMPLES // train_batch_size
return cfg.ExperimentConfig(
task=SimCLRPretrainTask(
model=SimCLRModel(
mode=simclr_model.PRETRAIN,
backbone_trainable=True,
input_size=[224, 224, 3],
backbone=backbones.Backbone(
type='resnet', resnet=backbones.ResNet(model_id=50)),
projection_head=ProjectionHead(
proj_output_dim=128, num_proj_layers=3, ft_proj_idx=1),
supervised_head=SupervisedHead(num_classes=1001),
norm_activation=common.NormActivation(
norm_momentum=0.9, norm_epsilon=1e-5, use_sync_bn=True)),
loss=ContrastiveLoss(),
evaluation=Evaluation(),
train_data=DataConfig(
parser=Parser(mode=simclr_model.PRETRAIN),
decoder=Decoder(decode_label=True),
input_path=os.path.join(IMAGENET_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size),
validation_data=DataConfig(
parser=Parser(mode=simclr_model.PRETRAIN),
decoder=Decoder(decode_label=True),
input_path=os.path.join(IMAGENET_INPUT_PATH_BASE, 'valid*'),
is_training=False,
global_batch_size=eval_batch_size),
),
trainer=cfg.TrainerConfig(
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
train_steps=500 * steps_per_epoch,
validation_steps=IMAGENET_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'lars',
'lars': {
'momentum':
0.9,
'weight_decay_rate':
0.000001,
'exclude_from_weight_decay': [
'batch_normalization', 'bias'
]
}
},
'learning_rate': {
'type': 'cosine',
'cosine': {
# 0.2 * BatchSize / 256
'initial_learning_rate': 0.2 * train_batch_size / 256,
# train_steps - warmup_steps
'decay_steps': 475 * steps_per_epoch
}
},
'warmup': {
'type': 'linear',
'linear': {
# 5% of total epochs
'warmup_steps': 25 * steps_per_epoch
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
])
def test_builder(self, backbone_type, decoder_type, input_size,
quantize_detection_head, quantize_detection_decoder):
num_classes = 2
input_specs = tf.keras.layers.InputSpec(
shape=[None, input_size[0], input_size[1], 3])
if backbone_type == 'spinenet_mobile':
backbone_config = backbones.Backbone(
type=backbone_type,
spinenet_mobile=backbones.SpineNetMobile(
model_id='49',
stochastic_depth_drop_rate=0.2,
min_level=3,
max_level=7,
use_keras_upsampling_2d=True))
elif backbone_type == 'mobilenet':
backbone_config = backbones.Backbone(type=backbone_type,
mobilenet=backbones.MobileNet(
model_id='MobileNetV2',
filter_size_scale=1.0))
else:
raise ValueError(
'backbone_type {} is not supported'.format(backbone_type))
if decoder_type == 'identity':
decoder_config = decoders.Decoder(type=decoder_type)
elif decoder_type == 'fpn':
decoder_config = decoders.Decoder(type=decoder_type,
fpn=decoders.FPN(
num_filters=128,
use_separable_conv=True,
use_keras_layer=True))
else:
raise ValueError(
'decoder_type {} is not supported'.format(decoder_type))
model_config = retinanet_cfg.RetinaNet(
num_classes=num_classes,
input_size=[input_size[0], input_size[1], 3],
backbone=backbone_config,
decoder=decoder_config,
head=retinanet_cfg.RetinaNetHead(attribute_heads=None,
use_separable_conv=True))
l2_regularizer = tf.keras.regularizers.l2(5e-5)
# Build the original float32 retinanet model.
model = factory.build_retinanet(input_specs=input_specs,
model_config=model_config,
l2_regularizer=l2_regularizer)
# Call the model with dummy input to build the head part.
dummpy_input = tf.zeros([1] + model_config.input_size)
model(dummpy_input, training=True)
# Build the QAT model from the original model with quantization config.
qat_model = qat_factory.build_qat_retinanet(
model=model,
quantization=common.Quantization(
quantize_detection_decoder=quantize_detection_decoder,
quantize_detection_head=quantize_detection_head),
model_config=model_config)
if quantize_detection_head:
# head become a RetinaNetHeadQuantized when we apply quantization.
self.assertIsInstance(
qat_model.head,
qat_dense_prediction_heads.RetinaNetHeadQuantized)
else:
# head is a RetinaNetHead if we don't apply quantization on head part.
self.assertIsInstance(qat_model.head,
dense_prediction_heads.RetinaNetHead)
self.assertNotIsInstance(
qat_model.head,
qat_dense_prediction_heads.RetinaNetHeadQuantized)
if decoder_type == 'FPN':
if quantize_detection_decoder:
# FPN decoder become a general keras functional model after applying
# quantization.
self.assertNotIsInstance(qat_model.decoder, fpn.FPN)
else:
self.assertIsInstance(qat_model.decoder, fpn.FPN)
def retinanet_spinenet_mobile_coco() -> cfg.ExperimentConfig:
"""COCO object detection with mobile RetinaNet."""
train_batch_size = 256
eval_batch_size = 8
steps_per_epoch = COCO_TRAIN_EXAMPLES // train_batch_size
input_size = 384
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='float32'),
task=RetinaNetTask(
annotation_file=os.path.join(COCO_INPUT_PATH_BASE,
'instances_val2017.json'),
model=RetinaNet(
backbone=backbones.Backbone(
type='spinenet_mobile',
spinenet_mobile=backbones.SpineNetMobile(
model_id='49',
stochastic_depth_drop_rate=0.2,
min_level=3,
max_level=7,
use_keras_upsampling_2d=False)),
decoder=decoders.Decoder(type='identity',
identity=decoders.Identity()),
head=RetinaNetHead(num_filters=48, use_separable_conv=True),
anchor=Anchor(anchor_size=3),
norm_activation=common.NormActivation(use_sync_bn=True,
activation='swish'),
num_classes=91,
input_size=[input_size, input_size, 3],
min_level=3,
max_level=7),
losses=Losses(l2_weight_decay=3e-5),
train_data=DataConfig(input_path=os.path.join(
COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size,
parser=Parser(aug_rand_hflip=True,
aug_scale_min=0.1,
aug_scale_max=2.0)),
validation_data=DataConfig(input_path=os.path.join(
COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
train_steps=600 * steps_per_epoch,
validation_steps=COCO_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'sgd',
'sgd': {
'momentum': 0.9
}
},
'learning_rate': {
'type': 'stepwise',
'stepwise': {
'boundaries':
[575 * steps_per_epoch, 590 * steps_per_epoch],
'values': [
0.32 * train_batch_size / 256.0,
0.032 * train_batch_size / 256.0,
0.0032 * train_batch_size / 256.0
],
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 2000,
'warmup_learning_rate': 0.0067
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None',
])
return config
def cascadercnn_spinenet_coco() -> cfg.ExperimentConfig:
"""COCO object detection with Cascade RCNN-RS with SpineNet backbone."""
steps_per_epoch = 463
coco_val_samples = 5000
train_batch_size = 256
eval_batch_size = 8
config = cfg.ExperimentConfig(
runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
task=MaskRCNNTask(
annotation_file=os.path.join(COCO_INPUT_PATH_BASE,
'instances_val2017.json'),
model=MaskRCNN(
backbone=backbones.Backbone(
type='spinenet',
spinenet=backbones.SpineNet(
model_id='49',
min_level=3,
max_level=7,
)),
decoder=decoders.Decoder(
type='identity', identity=decoders.Identity()),
roi_sampler=ROISampler(cascade_iou_thresholds=[0.6, 0.7]),
detection_head=DetectionHead(
class_agnostic_bbox_pred=True, cascade_class_ensemble=True),
anchor=Anchor(anchor_size=3),
norm_activation=common.NormActivation(
use_sync_bn=True, activation='swish'),
num_classes=91,
input_size=[640, 640, 3],
min_level=3,
max_level=7,
include_mask=True),
losses=Losses(l2_weight_decay=0.00004),
train_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size,
parser=Parser(
aug_rand_hflip=True, aug_scale_min=0.1, aug_scale_max=2.5)),
validation_data=DataConfig(
input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
is_training=False,
global_batch_size=eval_batch_size,
drop_remainder=False)),
trainer=cfg.TrainerConfig(
train_steps=steps_per_epoch * 500,
validation_steps=coco_val_samples // eval_batch_size,
validation_interval=steps_per_epoch,
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'sgd',
'sgd': {
'momentum': 0.9
}
},
'learning_rate': {
'type': 'stepwise',
'stepwise': {
'boundaries': [
steps_per_epoch * 475, steps_per_epoch * 490
],
'values': [0.32, 0.032, 0.0032],
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 2000,
'warmup_learning_rate': 0.0067
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None',
'task.model.min_level == task.model.backbone.spinenet.min_level',
'task.model.max_level == task.model.backbone.spinenet.max_level',
])
return config
def image_classification_imagenet_revnet() -> cfg.ExperimentConfig:
"""Returns a revnet config for image classification on imagenet."""
train_batch_size = 4096
eval_batch_size = 4096
steps_per_epoch = IMAGENET_TRAIN_EXAMPLES // train_batch_size
config = cfg.ExperimentConfig(
task=ImageClassificationTask(
model=ImageClassificationModel(
num_classes=1001,
input_size=[224, 224, 3],
backbone=backbones.Backbone(
type='revnet', revnet=backbones.RevNet(model_id=56)),
norm_activation=common.NormActivation(norm_momentum=0.9,
norm_epsilon=1e-5,
use_sync_bn=False),
add_head_batch_norm=True),
losses=Losses(l2_weight_decay=1e-4),
train_data=DataConfig(input_path=os.path.join(
IMAGENET_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size),
validation_data=DataConfig(input_path=os.path.join(
IMAGENET_INPUT_PATH_BASE, 'valid*'),
is_training=False,
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
train_steps=90 * steps_per_epoch,
validation_steps=IMAGENET_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'sgd',
'sgd': {
'momentum': 0.9
}
},
'learning_rate': {
'type': 'stepwise',
'stepwise': {
'boundaries': [
30 * steps_per_epoch, 60 * steps_per_epoch,
80 * steps_per_epoch
],
'values': [0.8, 0.08, 0.008, 0.0008]
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 5 * steps_per_epoch,
'warmup_learning_rate': 0
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
])
return config
def image_classification_imagenet_resnetrs() -> cfg.ExperimentConfig:
"""Image classification on imagenet with resnet-rs."""
train_batch_size = 4096
eval_batch_size = 4096
steps_per_epoch = IMAGENET_TRAIN_EXAMPLES // train_batch_size
config = cfg.ExperimentConfig(
task=ImageClassificationTask(
model=ImageClassificationModel(
num_classes=1001,
input_size=[160, 160, 3],
backbone=backbones.Backbone(
type='resnet',
resnet=backbones.ResNet(model_id=50,
stem_type='v1',
resnetd_shortcut=True,
replace_stem_max_pool=True,
se_ratio=0.25,
stochastic_depth_drop_rate=0.0)),
dropout_rate=0.25,
norm_activation=common.NormActivation(norm_momentum=0.0,
norm_epsilon=1e-5,
use_sync_bn=False,
activation='swish')),
losses=Losses(l2_weight_decay=4e-5, label_smoothing=0.1),
train_data=DataConfig(
input_path=os.path.join(IMAGENET_INPUT_PATH_BASE, 'train*'),
is_training=True,
global_batch_size=train_batch_size,
aug_type=common.Augmentation(
type='randaug', randaug=common.RandAugment(magnitude=10))),
validation_data=DataConfig(input_path=os.path.join(
IMAGENET_INPUT_PATH_BASE, 'valid*'),
is_training=False,
global_batch_size=eval_batch_size)),
trainer=cfg.TrainerConfig(
steps_per_loop=steps_per_epoch,
summary_interval=steps_per_epoch,
checkpoint_interval=steps_per_epoch,
train_steps=350 * steps_per_epoch,
validation_steps=IMAGENET_VAL_EXAMPLES // eval_batch_size,
validation_interval=steps_per_epoch,
optimizer_config=optimization.OptimizationConfig({
'optimizer': {
'type': 'sgd',
'sgd': {
'momentum': 0.9
}
},
'ema': {
'average_decay': 0.9999,
'trainable_weights_only': False,
},
'learning_rate': {
'type': 'cosine',
'cosine': {
'initial_learning_rate': 1.6,
'decay_steps': 350 * steps_per_epoch
}
},
'warmup': {
'type': 'linear',
'linear': {
'warmup_steps': 5 * steps_per_epoch,
'warmup_learning_rate': 0
}
}
})),
restrictions=[
'task.train_data.is_training != None',
'task.validation_data.is_training != None'
])
return config
