swin_large(pretrained: bool = False, overrides: object = {})Swin-Large backbone (Liu et al., 2021).
Builds the canonical Swin-L configuration: embed_dim=192,
depths=(2, 2, 18, 2), num_heads=(6, 12, 24, 48). Final
feature dimension is . Approximately
197M parameters. Pretrained variants are typically released
only with ImageNet-22k pretraining due to the model's capacity.
Model Size
Parameters
pretrainedbool= FalseIf
True, loads ImageNet-22k pretrained weights when
available. Defaults to False.**overridesobject= {}Keyword overrides on top of the canonical Swin-L config.
Returns
SwinTransformerA SwinTransformer backbone returning a flat
feature.
Notes
Swin-L reaches 87.3% top-1 on ImageNet-1k (384x384) after ImageNet-22k pretraining (Liu et al., 2021, Table 2).
Examples
>>> import lucid
>>> from lucid.models.vision.swin import swin_large
>>> model = swin_large()
>>> x = lucid.randn(1, 3, 224, 224)
>>> model.forward_features(x).shape
(1, 1536)