class

ResNeXtForImageClassification

extendsPretrainedModelClassificationHeadMixin

ResNeXtForImageClassification(config: ResNeXtConfig)

source edit

ResNeXt image classifier (backbone + GAP + linear head).

Combines a ResNeXt backbone with the standard ImageNet head: lucid.nn.AdaptiveAvgPool2d to pool every spatial location into a single 2048-dim feature, an optional lucid.nn.Dropout (controlled by config.dropout), and a lucid.nn.Linear projection to config.num_classes. When labels are supplied to forward, a cross-entropy loss is returned alongside the logits.

Parameters

configResNeXtConfig

Architecture spec. Use the *_cls factory functions (resnext_50_32x4d_cls, resnext_101_32x8d_cls, …) for paper-cited configurations.

Attributes

configResNeXtConfig

Stored copy of the config that built this model.

stem, maxpool, layer1, layer2, layer3, layer4

Same backbone components as on ResNeXt.

avgpoolnn.AdaptiveAvgPool2d

Global average pool to

1\times1

classifiernn.Module

Final classifier built by ClassificationHeadMixin._build_classifier; bare linear if dropout == 0.0, otherwise a lucid.nn.Sequential wrapping lucid.nn.Dropout and lucid.nn.Linear.

Notes

From Xie et al., "Aggregated Residual Transformations for Deep Neural Networks", CVPR 2017. Loss is the standard categorical cross-entropy

\mathcal{L} = -\frac{1}{N} \sum_{n=1}^{N} \log \operatorname{softmax}(\text{logits}_n)_{\,y_n}.

The headline ImageNet result is ResNeXt-101 (64x4d) reaching 20.4% top-1 error — meaningfully better than ResNet-152 at the same FLOP budget.

Examples

>>> import lucid
>>> from lucid.models.vision.resnext import resnext_50_32x4d_cls
>>> model = resnext_50_32x4d_cls()
>>> x = lucid.randn(2, 3, 224, 224)
>>> out = model(x)
>>> out.logits.shape
(2, 1000)

Used by 2

Constructors

dunder

init

→None

__init__(config: ResNeXtConfig)

source edit

Instance methods

forward

→ImageClassificationOutput

forward(x: Tensor, labels: Tensor | None = None)

source edit

>>> import lucid >>> from lucid.models.vision.resnext import resnext_50_32x4d_cls >>> model = resnext_50_32x4d_cls() >>> x = lucid.randn(2, 3, 224, 224) >>> out = model(x) >>> out.logits.shape (2, 1000)