Pretrained weight tags for lucid.models.swin_large_cls.
Ships the original Microsoft ImageNet-22k → ImageNet-1k finetune
(MS_IN22K_FT_IN1K), re-hosted via timm
(swin_large_patch4_window7_224.ms_in22k_ft_in1k) under
huggingface.co/lucid-dl/swin-large. 86.3% top-1 at 224×224 is
the figure published for this checkpoint tag; the paper does not
report a 224×224 number for Swin-L, and its Table 2 is COCO
detection, not classification.
Examples
>>> from lucid.models.vision.swin._weights import SwinLargeWeights
>>> list(SwinLargeWeights.__members__)
['MS_IN22K_FT_IN1K', 'DEFAULT']
DEFAULT is an alias rather than a fourth entry, so a bare
pretrained=True and the tag it resolves to cannot drift apart.
>>> SwinLargeWeights.DEFAULT is SwinLargeWeights.MS_IN22K_FT_IN1K
True
>>> SwinLargeWeights.MS_IN22K_FT_IN1K.num_classes
1000