data

InceptionNeXtConfig

extendsModelConfig

InceptionNeXtConfig(num_classes: int = 1000, in_channels: int = 3, depths: tuple[int, ...] = (3, 3, 9, 3), dims: tuple[int, ...] = (96, 192, 384, 768), band_kernel: int = 11, mlp_ratios: tuple[int, ...] = (4, 4, 4, 3))

source edit

Configuration dataclass for every InceptionNeXt variant.

InceptionNeXtConfig is an immutable container that fully specifies the architecture of an InceptionNeXt (Yu et al., 2024). InceptionNeXt is a drop-in replacement for the ConvNeXt block that factorizes the single large $7 \times 7$ depthwise convolution into four parallel Inception-style branches:

\mathrm{IDWConv}(x) = \mathrm{Concat}\bigl( x^{(1)},\; \mathrm{DW}_{3 \times 3}(x^{(2)}),\; \mathrm{DW}_{1 \times K}(x^{(3)}),\; \mathrm{DW}_{K \times 1}(x^{(4)})\bigr).

The remaining elements of the ConvNeXt block (BatchNorm, inverted- bottleneck Conv-MLP, layer scale, residual) are kept unchanged, so the Tiny / Small / Base variants reuse the same depth and width tables as ConvNeXt while running noticeably faster.

Parameters

num_classesint= 1000

Number of output classes for the classification head. Defaults to 1000 (ImageNet-1k).

in_channelsint= 3

Number of input image channels. Defaults to 3 (RGB).

depthstuple of int= (3, 3, 9, 3)

Number of MetaNeXt blocks in each of the four stages. Defaults to (3, 3, 9, 3) (InceptionNeXt-T).

dimstuple of int= (96, 192, 384, 768)

Channel width per stage. Defaults to (96, 192, 384, 768) (InceptionNeXt-T).

band_kernelint= 11

Kernel length

K

of the horizontal

1 \times K

and vertical

K \times 1

band depthwise convolutions inside the Inception token mixer. Defaults to 11.

mlp_ratiostuple of int= (4, 4, 4, 3)

Per-stage MLP expansion ratios. Defaults to (4, 4, 4, 3) — the reference recipe uses ratio 3 in the final stage to match the head's compute budget.

Attributes

model_typeClassVar[str]

Constant string "inception_next" used by the model registry.

Notes

Reference: Weihao Yu et al., "InceptionNeXt: When Inception Meets ConvNeXt", CVPR 2024, arXiv:2303.16900.

Examples

Build an InceptionNeXt-T configuration with a 100-class head:
>>> from lucid.models.vision.inception_next import InceptionNeXtConfig
>>> cfg = InceptionNeXtConfig(num_classes=100)
>>> cfg.depths, cfg.dims, cfg.band_kernel, cfg.num_classes
((3, 3, 9, 3), (96, 192, 384, 768), 11, 100)

Used by 3

Constructors

dunder

init

→None

__init__(num_classes: int = 1000, in_channels: int = 3, depths: tuple[int, ...] = (3, 3, 9, 3), dims: tuple[int, ...] = (96, 192, 384, 768), band_kernel: int = 11, mlp_ratios: tuple[int, ...] = (4, 4, 4, 3))

source edit

InceptionNeXtConfig(num_classes: int = 1000, in_channels: int = 3, depths: tuple[int, ...] = (3, 3, 9, 3), dims: tuple[int, ...] = (96, 192, 384, 768), band_kernel: int = 11, mlp_ratios: tuple[int, ...] = (4, 4, 4, 3))

Build an InceptionNeXt-T configuration with a 100-class head: >>> from lucid.models.vision.inception_next import InceptionNeXtConfig >>> cfg = InceptionNeXtConfig(num_classes=100) >>> cfg.depths, cfg.dims, cfg.band_kernel, cfg.num_classes ((3, 3, 9, 3), (96, 192, 384, 768), 11, 100)

__init__(num_classes: int = 1000, in_channels: int = 3, depths: tuple[int, ...] = (3, 3, 9, 3), dims: tuple[int, ...] = (96, 192, 384, 768), band_kernel: int = 11, mlp_ratios: tuple[int, ...] = (4, 4, 4, 3))