inception_next_base

→InceptionNeXt

inception_next_base(pretrained: bool = False, overrides: object = {})

source edit

InceptionNeXt-Base backbone (Yu et al., 2024).

Builds the canonical InceptionNeXt-B configuration: depths=(3, 3, 27, 3), dims=(128, 256, 512, 1024), band_kernel=11. Approximately 87M parameters — the widest of the paper's three main variants.

Model Size

Parameters

pretrainedbool= False

If True, loads ImageNet-1k pretrained weights when available in the model zoo. Defaults to False.

**overridesobject= {}

Keyword overrides applied on top of the canonical InceptionNeXt-B config.

Returns

InceptionNeXt

An InceptionNeXt backbone returning a flat $(B, 1024)$ feature.

Notes

InceptionNeXt-B reaches 84.0% top-1 on ImageNet-1k (Yu et al., 2024, Table 2). See arXiv:2303.16900.

Examples

>>> import lucid
>>> from lucid.models.vision.inception_next import inception_next_base
>>> model = inception_next_base()
>>> x = lucid.randn(1, 3, 224, 224)
>>> feat = model.forward_features(x)
>>> feat.shape
(1, 1024)

Used by 1

lucid.models.vision.inception_next

>>> import lucid >>> from lucid.models.vision.inception_next import inception_next_base >>> model = inception_next_base() >>> x = lucid.randn(1, 3, 224, 224) >>> feat = model.forward_features(x) >>> feat.shape (1, 1024)