se_resnet_18¶
Overview¶
The se_resnet_18 function constructs an SE-ResNet-18 model, a lightweight residual network augmented with Squeeze-and-Excitation (SE) blocks for adaptive recalibration of feature maps. It is suitable for image classification tasks.
Total Parameters: 11,778,592
Function Signature¶
@register_model
def se_resnet_18(num_classes: int = 1000, **kwargs) -> SENet:
Parameters¶
num_classes (int, optional): Number of output classes for the classification task. Default is 1000.
kwargs: Additional keyword arguments to customize the model.
Returns¶
SENet: An instance of the SE-ResNet-18 model.
Examples¶
Creating an SE-ResNet-18 model for 1000 classes:
model = se_resnet_18(num_classes=1000)
print(model)
Note
SE-ResNet-18 uses a configuration of [2, 2, 2, 2] for its layers.
Incorporates _SEResNetModule for SE operations.
Initializes weights internally unless specified otherwise through kwargs.