convnext_xlarge(pretrained: bool = False, overrides: object = {})ConvNeXt-XLarge backbone (Liu et al., 2022).
Builds the canonical ConvNeXt-XL configuration:
depths=(3, 3, 27, 3), dims=(256, 512, 1024, 2048). Final
feature width is 2048 — twice the ConvNeXt-B width. Approximately
350M parameters — the largest variant in the paper.
Model Size
Parameters
pretrainedbool= FalseIf
True, loads ImageNet-22k pretrained weights when
available. Defaults to False.**overridesobject= {}Keyword overrides on top of the canonical ConvNeXt-XL config.
Returns
ConvNeXtA ConvNeXt backbone returning a flat
feature.
Notes
ConvNeXt-XL reaches 87.0% top-1 on ImageNet-1k at 384x384 fine-tune resolution after ImageNet-22k pretraining (Liu et al., 2022, Table 11) — the headline result of the paper.
Examples
>>> import lucid
>>> from lucid.models.vision.convnext import convnext_xlarge
>>> model = convnext_xlarge()
>>> x = lucid.randn(1, 3, 224, 224)
>>> model.forward_features(x).shape
(1, 2048)