VGGNetConfig¶

class lucid.models.VGGNetConfig(conv_config: list[int | str], num_classes: int = 1000, in_channels: int = 3, dropout: float = 0.5, classifier_hidden_features: tuple[int, int] = (4096, 4096))¶

VGGNetConfig stores the architectural choices used by lucid.models.VGGNet. It defines the convolution and pooling schedule, the output class count, the input channel count, the classifier hidden widths, and the dropout probability used in the classifier head.

Class Signature¶

@dataclass
class VGGNetConfig:
    conv_config: list[int | str]
    num_classes: int = 1000
    in_channels: int = 3
    dropout: float = 0.5
    classifier_hidden_features: tuple[int, int] = (4096, 4096)

Parameters¶

conv_config (list[int | str]): Convolution/pooling schedule where positive integers denote convolution output channels and “M” denotes max pooling.
num_classes (int): Number of output classes.
in_channels (int): Number of channels in the input image tensor.
dropout (float): Dropout probability used in the classifier head.
classifier_hidden_features (tuple[int, int]): Hidden widths of the two classifier linear layers.

Validation¶

conv_config must not be empty and must include at least 1 conv layer.
Each conv_config entry must be a positive integer or “M”.
num_classes and in_channels must be greater than 0.
dropout must be in the range [0.0, 1.0).
classifier_hidden_features must contain exactly 2 positive integers.

Usage¶

import lucid.models as models

config = models.VGGNetConfig(
    conv_config=[64, "M", 128, "M", 256, 256, "M"],
    num_classes=10,
    in_channels=1,
    classifier_hidden_features=(512, 256),
    dropout=0.25,
)

model = models.VGGNet(config)