class

SKNetForImageClassification

extendsPretrainedModelClassificationHeadMixin

SKNetForImageClassification(config: SKNetConfig)

source edit

SK-ResNet with global-average-pooled linear classification head.

Combines an SKNet backbone with global average pooling and a linear projection to config.num_classes logits. When labels are supplied to forward, a cross-entropy loss is computed and returned alongside the logits.

Parameters

configSKNetConfig

Architecture spec. Use the *_cls factory functions (sk_resnet_18_cls through sk_resnext_50_32x4d_cls) for paper-cited configurations.

Attributes

configSKNetConfig

Stored copy of the config that built this model.

stem, maxpool, layer1, layer2, layer3, layer4

Same backbone components as on SKNet; see that class for shape semantics.

avgpoolnn.AdaptiveAvgPool2d

Global average pool collapsing the final feature map to 1 × 1.

classifiernn.Module

Built by ClassificationHeadMixin._build_classifier — a lucid.nn.Linear (or lucid.nn.Sequential with dropout) projecting from 512 (basic) or 2048 (bottleneck) channels to config.num_classes.

Notes

The classification flow is

\text{logits} = W \,\operatorname{GAP}(\mathrm{backbone}(x)) + b.

Examples

Run inference on a batch of 224×224 RGB images:
>>> import lucid
>>> from lucid.models.vision.sknet import sk_resnet_50_cls
>>> model = sk_resnet_50_cls()
>>> x = lucid.randn(4, 3, 224, 224)
>>> out = model(x)
>>> out.logits.shape
(4, 1000)

Used by 2

Constructors

dunder

init

→None

__init__(config: SKNetConfig)

source edit

Instance methods

forward

→ImageClassificationOutput

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

source edit

Run inference on a batch of 224×224 RGB images: >>> import lucid >>> from lucid.models.vision.sknet import sk_resnet_50_cls >>> model = sk_resnet_50_cls() >>> x = lucid.randn(4, 3, 224, 224) >>> out = model(x) >>> out.logits.shape (4, 1000)