XceptionConfig¶

class lucid.models.XceptionConfig(num_classes: int = 1000, in_channels: int = 3, stem_channels: tuple[int, int] | list[int] = (32, 64), entry_channels: tuple[int, int, int] | list[int] = (128, 256, 728), middle_channels: int = 728, middle_repeats: int = 8, exit_channels: tuple[int, int, int] | list[int] = (1024, 1536, 2048))¶

XceptionConfig stores the architectural settings used by lucid.models.Xception. It defines the classifier size, input channels, and the channel widths and repeat counts used by the Xception entry, middle, and exit flows.

Class Signature¶

@dataclass
class XceptionConfig:
    num_classes: int = 1000
    in_channels: int = 3
    stem_channels: tuple[int, int] | list[int] = (32, 64)
    entry_channels: tuple[int, int, int] | list[int] = (128, 256, 728)
    middle_channels: int = 728
    middle_repeats: int = 8
    exit_channels: tuple[int, int, int] | list[int] = (1024, 1536, 2048)

Parameters¶

num_classes (int): Number of output classes.
in_channels (int): Number of channels in the input image tensor.
stem_channels (tuple[int, int] | list[int]): Output widths of the two stem convolutions.
entry_channels (tuple[int, int, int] | list[int]): Output widths of the three entry-flow residual blocks.
middle_channels (int): Channel width used by the repeated middle-flow blocks.
middle_repeats (int): Number of repeated middle-flow residual blocks.
exit_channels (tuple[int, int, int] | list[int]): Channel widths used by the exit-flow block and the two final separable convolutions.

Validation¶

num_classes, in_channels, middle_channels, and middle_repeats must be greater than 0.
stem_channels must contain exactly two positive integers.
entry_channels and exit_channels must each contain exactly three positive integers.

Usage¶

import lucid.models as models

config = models.XceptionConfig(
    num_classes=10,
    in_channels=1,
    middle_repeats=8,
)

model = models.Xception(config)