class

ResNetForImageClassification

extendsPretrainedModelClassificationHeadMixin

ResNetForImageClassification(config: ResNetConfig)

source edit

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

Combines a ResNet backbone with the standard ImageNet classification head: an lucid.nn.AdaptiveAvgPool2d to pool every spatial location into a single feature vector, an optional lucid.nn.Dropout (controlled by config.dropout), and a lucid.nn.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

configResNetConfig

Architecture spec. Use the *_cls factory functions (resnet_18_cls, resnet_50_cls, …) to obtain a paper-cited configuration; pass a custom config to change num_classes or enable dropout / zero_init_residual.

Attributes

configResNetConfig

Stored copy of the config that built this model.

stem, maxpool, layer1, layer2, layer3, layer4

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

avgpoolnn.AdaptiveAvgPool2d

Global average pool collapsing the H/32 × W/32 feature map to 1 × 1.

classifiernn.Module

Built by ClassificationHeadMixin._build_classifier — either a bare lucid.nn.Linear (dropout == 0.0) or lucid.nn.Sequential wrapping lucid.nn.Dropout and lucid.nn.Linear.

Notes

The classification flow is

\text{logits} = W\,\operatorname{Drop}\! \left(\operatorname{GAP}(\,\mathrm{backbone}(x)\,)\right) + b,

where $\operatorname{GAP}$ is the global average pool and $W \in \mathbb{R}^{C \times D}$ with $D = \text{hidden\_sizes}[3] \times \text{expansion}$ . Loss is the standard categorical cross-entropy

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

computed only when labels is not None — useful for the common pattern of running inference without a label tensor.

Examples

Run inference on a batch of 224x224 RGB images:
>>> import lucid
>>> from lucid.models.vision.resnet import resnet_50_cls
>>> model = resnet_50_cls()
>>> x = lucid.randn(4, 3, 224, 224)
>>> out = model(x)
>>> out.logits.shape
(4, 1000)
>>> out.loss is None
True
Compute a loss given labels (e.g. during training):
>>> labels = lucid.tensor([0, 1, 2, 3], dtype=lucid.int64)
>>> out = model(x, labels=labels)
>>> out.loss.shape
()

Used by 2

Constructors

dunder

init

→None

__init__(config: ResNetConfig)

source edit

Initialise the module and validate the supplied config.

Parameters

configModelConfig

Must be an instance of the subclass's declared config_class.

Raises

TypeError

If config_class has not been set on the concrete subclass, or if config is not an instance of config_class.

Instance methods

forward

→ImageClassificationOutput

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

source edit

Run inference on a batch of 224x224 RGB images: >>> import lucid >>> from lucid.models.vision.resnet import resnet_50_cls >>> model = resnet_50_cls() >>> x = lucid.randn(4, 3, 224, 224) >>> out = model(x) >>> out.logits.shape (4, 1000) >>> out.loss is None True Compute a loss given labels (e.g. during training): >>> labels = lucid.tensor([0, 1, 2, 3], dtype=lucid.int64) >>> out = model(x, labels=labels) >>> out.loss.shape ()