resnext_101_32x8d

→ResNeXt

resnext_101_32x8d(pretrained: bool = False, overrides: object = {})

source edit

ResNeXt-101 (32x8d) feature-extracting backbone.

Builds a ResNeXt with the higher-capacity ResNeXt-101 (32x8d) topology: per-stage block counts (3, 4, 23, 3), cardinality $C = 32$ , width per group $d = 8$ (double the standard ResNeXt-101). Approximately 88.8 M parameters — the widest of the canonical ResNeXt variants, widely used as the backbone for ImageNet-pretrained downstream models (e.g. Facebook's Instagram-pretrained ig_resnext_101_32x8d).

Model Size

Parameters

pretrainedbool= False

Reserved for future pretrained-weight loading. Currently ignored.

**overridesobject= {}

Keyword overrides forwarded into ResNeXtConfig.

Returns

ResNeXt

Backbone with the ResNeXt-101 (32x8d) configuration applied (or with overrides merged on top of it).

Notes

See Xie et al., CVPR 2017. This widened variant is included in the reference-framework model zoo because of its strong transfer learning performance.

Examples

>>> import lucid
>>> from lucid.models.vision.resnext import resnext_101_32x8d
>>> model = resnext_101_32x8d()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape   # (B, 2048, 7, 7)
(1, 2048, 7, 7)

Used by 1

lucid.models.vision.resnext

>>> import lucid >>> from lucid.models.vision.resnext import resnext_101_32x8d >>> model = resnext_101_32x8d() >>> x = lucid.randn(1, 3, 224, 224) >>> out = model(x) >>> out.last_hidden_state.shape # (B, 2048, 7, 7) (1, 2048, 7, 7)