sparseml.keras.models.classification package

Submodules

sparseml.keras.models.classification.resnet module

Keras ResNet implementation. Further info on ResNet can be found in the paper here.

class sparseml.keras.models.classification.resnet.ResNetSection(num_blocks: int, out_channels: int, downsample: bool, proj_channels: int = - 1)[source]

Bases: object

Settings to describe how to put together a ResNet based architecture using user supplied configurations.

Parameters

  • num_blocks – the number of blocks to put in the section (ie Basic or Bottleneck blocks)

  • out_channels – the number of output channels from the section

  • downsample – True to apply stride 2 for downsampling of the input, False otherwise

  • proj_channels – The number of channels in the projection for a bottleneck block, if < 0 then uses basic

create(name: str, x_tens: tensorflow.python.framework.ops.Tensor, training: Union[bool, tensorflow.python.framework.ops.Tensor], kernel_initializer, bias_initializer, beta_initializer, gamma_initializer) → tensorflow.python.framework.ops.Tensor[source]

Create the section in the current graph and scope

Parameters

  • name – the name for the scope to create the section under

  • x_tens – The input tensor to the ResNet architecture

  • training – bool or Tensor to specify if the model should be run in training or inference mode

  • kernel_initializer – Initializer to use for the conv and fully connected kernels

  • bias_initializer – Initializer to use for the bias in the fully connected

  • beta_initializer – Initializer to use for the batch norm beta variables

  • gamma_initializer – Initializer to use for the batch norm gama variables

Returns

the output tensor from the section

sparseml.keras.models.classification.resnet.resnet101(pretrained_path: Optional[str] = None, pretrained: Union[bool, str] = False, pretrained_dataset: Optional[str] = None, inputs: tensorflow.python.framework.ops.Tensor = None, training: Union[bool, tensorflow.python.framework.ops.Tensor] = True, num_classes: int = 1000, class_type: str = None, kernel_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, bias_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, beta_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, gamma_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>) → tensorflow.python.keras.engine.training.Model[source]

Standard ResNet101 implementation; expected input shape is (B, 224, 224, 3)

Parameters

  • inputs – The input tensor to the ResNet architecture

  • training – bool or Tensor to specify if the model should be run in training or inference mode

  • num_classes – The number of classes to classify

  • class_type – One of [single, multi, None] to support multi class training. Default single. If None, then will not add the fully connected at the end.

  • kernel_initializer – Initializer to use for the conv and fully connected kernels

  • bias_initializer – Initializer to use for the bias in the fully connected

  • beta_initializer – Initializer to use for the batch norm beta variables

  • gamma_initializer – Initializer to use for the batch norm gama variables

  • pretrained_path – A path to the pretrained weights to load, if provided will override the pretrained param

  • pretrained – True to load the default pretrained weights, a string to load a specific pretrained weight (ex: base, optim, optim-perf), or False to not load any pretrained weights

  • pretrained_dataset – The dataset to load pretrained weights for (ex: imagenet, mnist, etc). If not supplied will default to the one preconfigured for the model.

Returns

the output tensor from the created graph

sparseml.keras.models.classification.resnet.resnet152(pretrained_path: Optional[str] = None, pretrained: Union[bool, str] = False, pretrained_dataset: Optional[str] = None, inputs: tensorflow.python.framework.ops.Tensor = None, training: Union[bool, tensorflow.python.framework.ops.Tensor] = True, num_classes: int = 1000, class_type: str = None, kernel_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, bias_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, beta_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, gamma_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>) → tensorflow.python.keras.engine.training.Model[source]

Standard ResNet152 implementation; expected input shape is (B, 224, 224, 3)

Parameters

  • inputs – The input tensor to the ResNet architecture

  • training – bool or Tensor to specify if the model should be run in training or inference mode

  • num_classes – The number of classes to classify

  • class_type – One of [single, multi, None] to support multi class training. Default single. If None, then will not add the fully connected at the end.

  • kernel_initializer – Initializer to use for the conv and fully connected kernels

  • bias_initializer – Initializer to use for the bias in the fully connected

  • beta_initializer – Initializer to use for the batch norm beta variables

  • gamma_initializer – Initializer to use for the batch norm gama variables

  • pretrained_path – A path to the pretrained weights to load, if provided will override the pretrained param

  • pretrained – True to load the default pretrained weights, a string to load a specific pretrained weight (ex: base, optim, optim-perf), or False to not load any pretrained weights

  • pretrained_dataset – The dataset to load pretrained weights for (ex: imagenet, mnist, etc). If not supplied will default to the one preconfigured for the model.

Returns

the output tensor from the created graph

sparseml.keras.models.classification.resnet.resnet50(pretrained_path: Optional[str] = None, pretrained: Union[bool, str] = False, pretrained_dataset: Optional[str] = None, inputs: tensorflow.python.framework.ops.Tensor = None, training: Union[bool, tensorflow.python.framework.ops.Tensor] = True, num_classes: int = 1000, class_type: str = None, kernel_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, bias_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, beta_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>, gamma_initializer=<tensorflow.python.ops.init_ops_v2.GlorotUniform object>) → tensorflow.python.keras.engine.training.Model[source]

Standard ResNet50 implementation; expected input shape is (B, 224, 224, 3)

Parameters

  • inputs – The input tensor to the ResNet architecture

  • training – bool or Tensor to specify if the model should be run in training or inference mode

  • num_classes – The number of classes to classify

  • class_type – One of [single, multi, None] to support multi class training. Default single. If None, then will not add the fully connected at the end.

  • kernel_initializer – Initializer to use for the conv and fully connected kernels

  • bias_initializer – Initializer to use for the bias in the fully connected

  • beta_initializer – Initializer to use for the batch norm beta variables

  • gamma_initializer – Initializer to use for the batch norm gama variables

  • pretrained_path – A path to the pretrained weights to load, if provided will override the pretrained param

  • pretrained – True to load the default pretrained weights, a string to load a specific pretrained weight (ex: base, optim, optim-perf), or False to not load any pretrained weights

  • pretrained_dataset – The dataset to load pretrained weights for (ex: imagenet, mnist, etc). If not supplied will default to the one preconfigured for the model.

Returns

the output tensor from the created graph

sparseml.keras.models.classification.resnet.resnet_const(x_tens: tensorflow.python.framework.ops.Tensor, training: Union[bool, tensorflow.python.framework.ops.Tensor], sec_settings: List[sparseml.keras.models.classification.resnet.ResNetSection], num_classes: int, class_type: str, kernel_initializer, bias_initializer, beta_initializer, gamma_initializer) → tensorflow.python.keras.engine.training.Model[source]

Graph constructor for ResNet implementation.

Parameters

  • x_tens – The input tensor to the ResNet architecture

  • training – bool or Tensor to specify if the model should be run in training or inference mode

  • sec_settings – The settings for each section in the ResNet modoel

  • num_classes – The number of classes to classify

  • class_type – One of [single, multi, None] to support multi class training. Default single. If None, then will not add the fully connected at the end.

  • kernel_initializer – Initializer to use for the conv and fully connected kernels

  • bias_initializer – Initializer to use for the bias in the fully connected

  • beta_initializer – Initializer to use for the batch norm beta variables

  • gamma_initializer – Initializer to use for the batch norm gama variables

Returns

the output tensor from the created graph

Module contents