Swin-Small

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

Lucid port of torchvision/Swin_S_Weights.IMAGENET1K_V1, converted to Lucid-native safetensors.

Available weights

Tag	acc@1	acc@5	Params	GFLOPs	Size	Source
IMAGENET1K_V1 (default)	83.196	96.36	49.6M	8.741	189.26 MB	torchvision

Usage

import lucid.models as models
from lucid.models.weights import SwinSmallWeights

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

# explicit tag (enum or string)
model = models.swin_small_cls(weights=SwinSmallWeights.IMAGENET1K_V1)
model = models.swin_small_cls(pretrained="IMAGENET1K_V1")

# preprocessing travels with the weights
weights = SwinSmallWeights.IMAGENET1K_V1
preprocess = weights.transforms()
logits = model(preprocess(image)[None]).logits

Conversion

Converted from torchvision/Swin_S_Weights.IMAGENET1K_V1 via python -m tools.convert_weights swin_small --tag IMAGENET1K_V1. 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}
}