data

VGGConfig

extendsModelConfig

VGGConfig(num_classes: int = 1000, in_channels: int = 3, arch: tuple[int, ...] = (2, 2, 3, 3, 3), batch_norm: bool = False, dropout: float = 0.5)

Configuration for all VGG variants (A/B/D/E ≡ 11/13/16/19).

arch encodes the per-block conv counts:

batch_norm enables BatchNorm after each Conv+ReLU pair (VGG-BN). dropout applies to the two 4096-dim FC layers (0.5 in the paper).

Constructors

dunder

→None

__init__(num_classes: int = 1000, in_channels: int = 3, arch: tuple[int, ...] = (2, 2, 3, 3, 3), batch_norm: bool = False, dropout: float = 0.5)