Pretrained weights for lucid.models.convnext_xlarge_cls.
Currently ships FB_IN22K_FT_IN1K — Facebook AI Research's
ImageNet-22k pretraining → ImageNet-1k finetune checkpoint
(paper Table 11, acc@1 ≈ 87.0%), sourced via timm's
convnext_xlarge.fb_in22k_ft_in1k and re-hosted under
huggingface.co/lucid-dl/convnext-xlarge. reference_vision does not
publish a 1k-class ConvNeXt-XLarge head.
Examples
>>> from lucid.models.vision.convnext._weights import ConvNeXtXLargeWeights
>>> list(ConvNeXtXLargeWeights.__members__)
['FB_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.
>>> ConvNeXtXLargeWeights.DEFAULT is ConvNeXtXLargeWeights.FB_IN22K_FT_IN1K
True
>>> ConvNeXtXLargeWeights.FB_IN22K_FT_IN1K.num_classes
1000