Using Templates

Parametric implementations of standard architectures.

Available Templates

ResNet: 18, 34, 50, 101, 152 layers
VGG: 11, 13, 16, 19 layers
MobileNet: V1, V2, V3 (Small/Large)
DenseNet: 121, 169, 201, 264 layers
EfficientNet: B0-B7 variants

Customization

Templates support method chaining for deep customization:

from torchvision_customizer import Template

model = (Template.resnet(layers=50)
         .replace_activation('swish')
         .replace_norm('group')
         .add_attention('cbam')
         .build(num_classes=10))