Keras
Keras 3 API documentation
KerasCV
Models
Backbones
MobileNetV3 backbones

MobileNetV3 backbones

source

MobileNetV3Backbone class 

keras_cv.models.MobileNetV3Backbone(
    stackwise_expansion,
    stackwise_filters,
    stackwise_kernel_size,
    stackwise_stride,
    stackwise_se_ratio,
    stackwise_activation,
    include_rescaling,
    input_shape=(None, None, 3),
    input_tensor=None,
    alpha=1.0,
    **kwargs
)

Instantiates the MobileNetV3 architecture.

References

For transfer learning use cases, make sure to read the guide to transfer learning & fine-tuning.

Arguments

  • stackwise_expansion: list of ints or floats, the expansion ratio for each inverted residual block in the model.
  • stackwise_filters: list of ints, number of filters for each inverted residual block in the model.
  • stackwise_stride: list of ints, stride length for each inverted residual block in the model.
  • include_rescaling: bool, whether to rescale the inputs. If set to True, inputs will be passed through a Rescaling(scale=1 / 255) layer.
  • input_shape: optional shape tuple, defaults to (None, None, 3).
  • input_tensor: optional Keras tensor (i.e., output of layers.Input()) to use as image input for the model.
  • alpha: float, controls the width of the network. This is known as the depth multiplier in the MobileNetV3 paper, but the name is kept for consistency with MobileNetV1 in Keras.
    • If alpha < 1.0, proportionally decreases the number of filters in each layer.
    • If alpha > 1.0, proportionally increases the number of filters in each layer.
    • If alpha = 1, default number of filters from the paper are used at each layer.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone with a custom config
model = MobileNetV3Backbone(
    stackwise_expansion=[1, 72.0 / 16, 88.0 / 24, 4, 6, 6, 3, 3, 6, 6, 6],
    stackwise_filters=[16, 24, 24, 40, 40, 40, 48, 48, 96, 96, 96],
    stackwise_kernel_size=[3, 3, 3, 5, 5, 5, 5, 5, 5, 5, 5],
    stackwise_stride=[2, 2, 1, 2, 1, 1, 1, 1, 2, 1, 1],
    stackwise_se_ratio=[0.25, None, None, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25],
    stackwise_activation=["relu", "relu", "relu", "hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish"],
    include_rescaling=False,
)
output = model(input_data)

source

from_preset method 

MobileNetV3Backbone.from_preset()

Instantiate MobileNetV3Backbone model from preset config and weights.

Arguments

  • preset: string. Must be one of “mobilenet_v3_small”, “mobilenet_v3_large”, “mobilenet_v3_large_imagenet”, “mobilenet_v3_small_imagenet”. If looking for a preset with pretrained weights, choose one of “mobilenet_v3_large_imagenet”, “mobilenet_v3_small_imagenet”.
  • load_weights: Whether to load pre-trained weights into model. Defaults to None, which follows whether the preset has pretrained weights available.

Examples

# Load architecture and weights from preset
model = keras_cv.models.MobileNetV3Backbone.from_preset(
    "mobilenet_v3_large_imagenet",
)
# Load randomly initialized model from preset architecture with weights
model = keras_cv.models.MobileNetV3Backbone.from_preset(
    "mobilenet_v3_large_imagenet",
    load_weights=False,
Preset nameParametersDescription
mobilenet_v3_small933.50KMobileNetV3 model with 14 layers where the batch normalization and hard-swish activation are applied after the convolution layers.
mobilenet_v3_large2.99MMobileNetV3 model with 28 layers where the batch normalization and hard-swish activation are applied after the convolution layers.
mobilenet_v3_large_imagenet2.99MMobileNetV3 model with 28 layers where the batch normalization and hard-swish activation are applied after the convolution layers. Pre-trained on the ImageNet 2012 classification task.
mobilenet_v3_small_imagenet933.50KMobileNetV3 model with 14 layers where the batch normalization and hard-swish activation are applied after the convolution layers. Pre-trained on the ImageNet 2012 classification task.

source

MobileNetV3SmallBackbone class 

keras_cv.models.MobileNetV3SmallBackbone(
    stackwise_expansion,
    stackwise_filters,
    stackwise_kernel_size,
    stackwise_stride,
    stackwise_se_ratio,
    stackwise_activation,
    include_rescaling,
    input_shape=(None, None, 3),
    input_tensor=None,
    alpha=1.0,
    **kwargs
)

MobileNetV3Backbone model with 14 layers.

References

For transfer learning use cases, make sure to read the guide to transfer learning & fine-tuning.

Arguments

  • include_rescaling: bool, whether to rescale the inputs. If set to True, inputs will be passed through a Rescaling(scale=1 / 255) layer. Defaults to True.
  • input_shape: optional shape tuple, defaults to (None, None, 3).
  • input_tensor: optional Keras tensor (i.e., output of layers.Input()) to use as image input for the model.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = MobileNetV3SmallBackbone()
output = model(input_data)

source

MobileNetV3LargeBackbone class 

keras_cv.models.MobileNetV3LargeBackbone(
    stackwise_expansion,
    stackwise_filters,
    stackwise_kernel_size,
    stackwise_stride,
    stackwise_se_ratio,
    stackwise_activation,
    include_rescaling,
    input_shape=(None, None, 3),
    input_tensor=None,
    alpha=1.0,
    **kwargs
)

MobileNetV3Backbone model with 28 layers.

References

For transfer learning use cases, make sure to read the guide to transfer learning & fine-tuning.

Arguments

  • include_rescaling: bool, whether to rescale the inputs. If set to True, inputs will be passed through a Rescaling(scale=1 / 255) layer. Defaults to True.
  • input_shape: optional shape tuple, defaults to (None, None, 3).
  • input_tensor: optional Keras tensor (i.e., output of layers.Input()) to use as image input for the model.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = MobileNetV3LargeBackbone()
output = model(input_data)