ResNetConfig¶

class lucid.models.ResNetConfig(block: Literal['basic', 'bottleneck', 'preact_bottleneck'] | Type[lucid.nn.module.Module], layers: tuple[int, int, int, int] | list[int], num_classes: int = 1000, in_channels: int = 3, stem_width: int = 64, stem_type: Literal['deep'] | None = None, avg_down: bool = False, channels: tuple[int, int, int, int] | list[int] = (64, 128, 256, 512), block_args: dict[str, typing.Any] = <factory>)¶

ResNetConfig stores the structural choices used by lucid.models.ResNet. It describes the residual block family, per-stage depths, stem settings, classifier size, and any extra block keyword arguments required by derived ResNet variants.

Class Signature¶

@dataclass
class ResNetConfig:
    block: Literal["basic", "bottleneck", "preact_bottleneck"] | Type[nn.Module]
    layers: tuple[int, int, int, int] | list[int]
    num_classes: int = 1000
    in_channels: int = 3
    stem_width: int = 64
    stem_type: Literal["deep"] | None = None
    avg_down: bool = False
    channels: tuple[int, int, int, int] | list[int] = (64, 128, 256, 512)
    block_args: dict[str, Any] = field(default_factory=dict)

Parameters¶

block (Literal[“basic”, “bottleneck”, “preact_bottleneck”] | Type[nn.Module]): Residual block selection. Built-in block names cover the standard ResNet families, and custom block classes are also accepted when they inherit from nn.Module and define an expansion attribute.
layers (tuple[int, int, int, int] | list[int]): Number of residual blocks in each of the four stages.
num_classes (int): Number of output classes.
in_channels (int): Number of channels in the input image tensor.
stem_width (int): Width parameter used by the deep stem variant.
stem_type (Literal[“deep”] | None): Stem style. None uses the classic single 7x7 convolution, while “deep” uses a three-layer stem.
avg_down (bool): Whether projection shortcuts should downsample using average pooling before the 1x1 projection.
channels (tuple[int, int, int, int] | list[int]): Output width of each residual stage.
block_args (dict[str, Any]): Extra keyword arguments forwarded to each residual block instance.

Validation¶

block must be one of the built-in block names or a custom nn.Module subclass.
Custom block classes must define an expansion attribute.
layers and channels must each contain exactly four positive integers.
num_classes, in_channels, and stem_width must be greater than 0.
stem_type must be None or “deep”.
block_args must be a dictionary.

Usage¶

import lucid.models as models

config = models.ResNetConfig(
    block="bottleneck",
    layers=[3, 4, 6, 3],
    num_classes=10,
    in_channels=1,
    stem_type="deep",
    stem_width=32,
    avg_down=True,
)

model = models.ResNet(config)