wide_resnet_101¶
The wide_resnet_101 function is a registered model that creates a Wide ResNet-101 architecture. This is a variant of the ResNet-101 model with a wider bottleneck configuration, often used to enhance the representational capacity of the network.
Total Parameters: 126,886,696
Function Signature¶
def wide_resnet_101(num_classes: int = 1000, **kwargs) -> ResNet
Parameters¶
num_classes (int, optional): The number of output classes for the final fully connected layer. Default is 1000, suitable for ImageNet classification.
kwargs (dict, optional): Additional keyword arguments passed to the ResNet constructor. These can include options for customization such as initialization or model-specific settings.
Returns¶
ResNet: A Wide ResNet-101 model configured with the specified number of output classes and additional options.
Architecture Details¶
Wide ResNet-101 is characterized by its wider bottleneck layers compared to the standard ResNet-101. Specifically, the base_width parameter determines the width of the bottleneck layers, with a default value of 128 in this implementation.
Note
The architecture comprises 4 stages with layers [3, 4, 23, 3], corresponding to the number of residual blocks in each stage.
Examples¶
Creating a Wide ResNet-101 model for ImageNet classification:
import lucid.models as models
# Create the model
model = models.wide_resnet_101(num_classes=1000)
# Print model summary
models.summarize(model)