Keras
Keras 3 API documentation
KerasCV
Models
Backbones
ResNetV2 backbones

ResNetV2 backbones

source

ResNetV2Backbone class 

keras_cv.models.ResNetV2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

Instantiates the ResNetV2 architecture.

Reference

The difference in Resnet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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

Arguments

  • stackwise_filters: list of ints, number of filters for each stack in the model.
  • stackwise_blocks: list of ints, number of blocks for each stack in the model.
  • stackwise_strides: list of ints, stride for each stack in the model.
  • include_rescaling: bool, whether to rescale the inputs. If set to True, inputs will be passed through a Rescaling(1/255.0) layer.
  • stackwise_dilations: list of ints, dilation for each stack in the model. If None (default), dilation will not be used.
  • 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.
  • block_type: string, one of “basic_block” or “block”. The block type to stack. Use “basic_block” for smaller models like ResNet18 and ResNet34.

Examples

input_data = tf.ones(shape=(8, 224, 224, 3))
# Pretrained backbone
model = keras_cv.models.ResNetV2Backbone.from_preset("resnet50_v2_imagenet")
output = model(input_data)
# Randomly initialized backbone with a custom config
model = ResNetV2Backbone(
    stackwise_filters=[64, 128, 256, 512],
    stackwise_blocks=[2, 2, 2, 2],
    stackwise_strides=[1, 2, 2, 2],
    include_rescaling=False,
)
output = model(input_data)

source

from_preset method 

ResNetV2Backbone.from_preset()

Instantiate ResNetV2Backbone model from preset config and weights.

Arguments

  • preset: string. Must be one of “resnet18_v2”, “resnet34_v2”, “resnet50_v2”, “resnet101_v2”, “resnet152_v2”, “resnet50_v2_imagenet”. If looking for a preset with pretrained weights, choose one of “resnet50_v2_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.ResNetV2Backbone.from_preset(
    "resnet50_v2_imagenet",
)
# Load randomly initialized model from preset architecture with weights
model = keras_cv.models.ResNetV2Backbone.from_preset(
    "resnet50_v2_imagenet",
    load_weights=False,
Preset nameParametersDescription
resnet18_v211.18MResNet model with 18 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style).
resnet34_v221.30MResNet model with 34 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style).
resnet50_v223.56MResNet model with 50 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style).
resnet101_v242.63MResNet model with 101 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style).
resnet152_v258.33MResNet model with 152 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style).
resnet50_v2_imagenet23.56MResNet model with 50 layers where the batch normalization and ReLU activation precede the convolution layers (v2 style). Trained on Imagenet 2012 classification task.

source

ResNet18V2Backbone class 

keras_cv.models.ResNet18V2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

ResNetV2Backbone model with 18 layers.

Reference

The difference in ResNet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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(1/255.0) 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.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = ResNet18V2Backbone()
output = model(input_data)

source

ResNet34V2Backbone class 

keras_cv.models.ResNet34V2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

ResNetV2Backbone model with 34 layers.

Reference

The difference in ResNet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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(1/255.0) 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.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = ResNet34V2Backbone()
output = model(input_data)

source

ResNet50V2Backbone class 

keras_cv.models.ResNet50V2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

ResNetV2Backbone model with 50 layers.

Reference

The difference in ResNet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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(1/255.0) 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.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = ResNet50V2Backbone()
output = model(input_data)

source

ResNet101V2Backbone class 

keras_cv.models.ResNet101V2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

ResNetV2Backbone model with 101 layers.

Reference

The difference in ResNet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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(1/255.0) 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.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = ResNet101V2Backbone()
output = model(input_data)

source

ResNet152V2Backbone class 

keras_cv.models.ResNet152V2Backbone(
    stackwise_filters,
    stackwise_blocks,
    stackwise_strides,
    include_rescaling,
    stackwise_dilations=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    block_type="block",
    **kwargs
)

ResNetV2Backbone model with 152 layers.

Reference

The difference in ResNet and ResNetV2 rests in the structure of their individual building blocks. In ResNetV2, the batch normalization and ReLU activation precede the convolution layers, as opposed to ResNetV1 where the batch normalization and ReLU activation are applied after the convolution layers.

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(1/255.0) 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.

Example

input_data = tf.ones(shape=(8, 224, 224, 3))
# Randomly initialized backbone
model = ResNet152V2Backbone()
output = model(input_data)