Swin-Large

Liu et al., 2021 — Swin Transformer: Hierarchical Vision Transformer using Shifted Windows (arXiv:2103.14030)

Lucid port of timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k, converted to Lucid-native safetensors.

Available weights

Tag	acc@1	acc@5	Params	GFLOPs	Size	Source
MS_IN22K_FT_IN1K (default)	86.32	97.89	196.5M	—	749.74 MB	timm

Usage

import lucid.models as models
from lucid.models.weights import SwinLargeWeights

# default tag
model = models.swin_large_cls(pretrained=True)

# explicit tag (enum or string)
model = models.swin_large_cls(weights=SwinLargeWeights.MS_IN22K_FT_IN1K)
model = models.swin_large_cls(pretrained="MS_IN22K_FT_IN1K")

# preprocessing travels with the weights
weights = SwinLargeWeights.MS_IN22K_FT_IN1K
preprocess = weights.transforms()
logits = model(preprocess(image)[None]).logits

Conversion

Converted from timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k via python -m tools.convert_weights swin_large --tag MS_IN22K_FT_IN1K. Key mapping + numerical parity verified against the source.

License

mit — inherited from the original weights.

Citation

@inproceedings{liu2021swin,
  title={Swin Transformer: Hierarchical Vision Transformer using Shifted Windows},
  author={Liu, Ze and Lin, Yutong and Cao, Yue and Hu, Han and Wei, Yixuan and Zhang, Zheng and Lin, Stephen and Guo, Baining},
  booktitle={ICCV}, year={2021}
}