lucid-dl/resnext-101-32x8d

ResNeXt-101 (32x8d)

Xie et al., 2017 — Aggregated Residual Transformations for Deep Neural Networks (arXiv:1611.05431)

Lucid port of torchvision/ResNeXt101_32X8D_Weights.IMAGENET1K_V2, converted to Lucid-native safetensors.

Available weights

Tag	acc@1	acc@5	Params	GFLOPs	Size	Source
IMAGENET1K_V2 (default)	82.834	96.228	88.8M	16.414	339.55 MB	torchvision

Usage

import lucid.models as models
from lucid.models.weights import ResNeXt101_32x8dWeights

# default tag
model = models.resnext_101_32x8d_cls(pretrained=True)

# explicit tag (enum or string)
model = models.resnext_101_32x8d_cls(weights=ResNeXt101_32x8dWeights.IMAGENET1K_V2)
model = models.resnext_101_32x8d_cls(pretrained="IMAGENET1K_V2")

# preprocessing travels with the weights
weights = ResNeXt101_32x8dWeights.IMAGENET1K_V2
preprocess = weights.transforms()
logits = model(preprocess(image)[None]).logits

Conversion

Converted from torchvision/ResNeXt101_32X8D_Weights.IMAGENET1K_V2 via python -m tools.convert_weights resnext_101_32x8d --tag IMAGENET1K_V2. Key mapping + numerical parity verified against the source.

License

bsd-3-clause — inherited from the original weights.

Citation

@inproceedings{xie2017aggregated,
  title={Aggregated Residual Transformations for Deep Neural Networks},
  author={Xie, Saining and Girshick, Ross and Doll{\'a}r, Piotr and Tu, Zhuowen and He, Kaiming},
  booktitle={CVPR}, year={2017}
}