resnext_101_32x4d_cls(pretrained: bool | str = False, weights: ResNeXt101_32x4dWeights | None = None, overrides: object = {})ResNeXt-101 (32x4d) image classifier (backbone + GAP + linear head).
Builds a ResNeXtForImageClassification with the
paper-cited ResNeXt-101 (32x4d) topology and a
lucid.nn.Linear classifier projecting 2048 →
config.num_classes. Approximately 44.2 M parameters; the
distributed Gluon GLUON_IN1K checkpoint reaches 80.342%
ImageNet-1k top-1.
Model Size
Parameters
pretrainedbool or str= FalsePretrained-weight selector.
False → random init; True
→ the DEFAULT tag
(ResNeXt101_32x4dWeights.GLUON_IN1K); a tag string
(e.g. "GLUON_IN1K") → that specific checkpoint. Mutually
exclusive with weights (which wins if both are given).Explicit weights enum member, e.g.
ResNeXt101_32x4dWeights.GLUON_IN1K. Takes precedence over
pretrained.**overridesobject= {}Keyword overrides forwarded into
ResNeXtConfig. Note:
overriding num_classes away from the checkpoint's class
count makes pretrained loading fail the strict key/shape check.Returns
ResNeXtForImageClassificationClassifier with the ResNeXt-101 (32x4d) configuration applied
(or with overrides merged on top of it), optionally
initialised from pretrained weights.
Notes
Pretrained weights are converted from the timm Gluon checkpoint
resnext101_32x4d.gluon_in1k and hosted on the Hugging Face Hub
under lucid-dl/resnext-101-32x4d. The Gluon preset uses bicubic
interpolation with a 0.875 crop_pct (256 resize → 224 crop).
Examples
>>> import lucid
>>> from lucid.models.vision.resnext import resnext_101_32x4d_cls
>>> model = resnext_101_32x4d_cls()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.logits.shape
(1, 1000)
Load ImageNet-pretrained weights:
>>> model = resnext_101_32x4d_cls(pretrained=True)
>>> from lucid.models.vision.resnext import ResNeXt101_32x4dWeights
>>> model = resnext_101_32x4d_cls(weights=ResNeXt101_32x4dWeights.GLUON_IN1K)