MobileNet v3

applications.mobilenet_v3.MobileNetV3Small

applications.mobilenet_v3.MobileNetV3Small(
    include_top=True, weights='hasc', input_shape=None, pooling=None, classes=6, classifier_activation='softmax',
    alpha=1.0, minimalistic=False
)

Reference: - Searching for MobileNetV3 (ICCV 2019)

By default, it loads weights pre-trained on HASC. Check 'weights' for other options.

The default input size for this model is 768 (256 * 3).

Arguments

• include_top : whether to include the fully-connected layer at the top of the network.
• weights : one of 'None' (he_normal initialization), 'hasc' (pre-training on HASC), or the path to the weights file to be loaded.
• input_shape : optional shape tuple, default (768, 1) (with channels_last data format).
• pooling : optional pooling mode for feature extraction when include_top is False.
  • None means that the output of the model will be applied to the 3D tensor output of the last convolutional block.
  • avg means that global average pooling will be applied to the output of the last convolutioinal block, and thus the output of the model will be a 2D tensor.
  • max means that global max pooling will be applied.
• classes : optional number of classes to classify images into, only to be specified if include_top is True, and if no weights argument is specified, default 6.
• classifier_activation : A str or callable. The activation function to use on the "top" layer. Ignored unless include_top=True. Set classifier_activation=None to return the logits of the "top" layer, default 'softmax'.
• alpha : Float between 0 and 1. controls the width of the network. This is known as the width multiplier in the MobileNetV3 paper.
• minimalistic: Bool. In addition to large and small models this module also contains so-called minimalistic models, these models have the same per-layer dimensions characteristic as MobilenetV3 however, they don't utilize any of the advanced blocks (squeeze-and-excite units, hard-swish, and 5x5 convolutions). While these models are less efficient on CPU, they are much more performant on GPU/DSP.

Returns

• tensorflow.keras.Model instance.

applications.mobilenet_v3.MobileNetV3Large

applications.mobilenet_v3.MobileNetV3Large(
    include_top=True, weights='hasc', input_shape=None, pooling=None, classes=6, classifier_activation='softmax',
    alpha=1.0, minimalistic=False
)

Reference: - Searching for MobileNetV3 (ICCV 2019)

By default, it loads weights pre-trained on HASC. Check 'weights' for other options.

The default input size for this model is 768 (256 * 3).

Arguments

• include_top : whether to include the fully-connected layer at the top of the network.
• weights : one of 'None' (he_normal initialization), 'hasc' (pre-training on HASC), or the path to the weights file to be loaded.
• input_shape : optional shape tuple, default (768, 1) (with channels_last data format).
• pooling : optional pooling mode for feature extraction when include_top is False.
  • None means that the output of the model will be applied to the 3D tensor output of the last convolutional block.
  • avg means that global average pooling will be applied to the output of the last convolutioinal block, and thus the output of the model will be a 2D tensor.
  • max means that global max pooling will be applied.
• classes : optional number of classes to classify images into, only to be specified if include_top is True, and if no weights argument is specified, default 6.
• classifier_activation : A str or callable. The activation function to use on the "top" layer. Ignored unless include_top=True. Set classifier_activation=None to return the logits of the "top" layer, default 'softmax'.
• alpha : Float between 0 and 1. controls the width of the network. This is known as the width multiplier in the MobileNetV3 paper.
• minimalistic: Bool. In addition to large and small models this module also contains so-called minimalistic models, these models have the same per-layer dimensions characteristic as MobilenetV3 however, they don't utilize any of the advanced blocks (squeeze-and-excite units, hard-swish, and 5x5 convolutions). While these models are less efficient on CPU, they are much more performant on GPU/DSP.

Returns

• tensorflow.keras.Model instance.