class

MobileNetV2ForImageClassification

extendsPretrainedModelClassificationHeadMixin

MobileNetV2ForImageClassification(config: MobileNetV2Config)

source edit

MobileNet v2 with global-average-pooled linear classification head.

Combines a MobileNetV2 backbone with the standard ImageNet classification head: an lucid.nn.AdaptiveAvgPool2d to pool every spatial location into a single feature vector, a lucid.nn.Dropout (probability config.dropout, default 0.2), 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

configMobileNetV2Config

Architecture spec. Use the *_cls factory functions (mobilenet_v2_cls, mobilenet_v2_075_cls) for paper-cited configurations.

Attributes

configMobileNetV2Config

Stored copy of the config that built this model.

featuresnn.Sequential

Same inverted-residual stack as on MobileNetV2.

avgpoolnn.AdaptiveAvgPool2d

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

dropnn.Dropout

Dropout layer (probability config.dropout) applied before the linear classifier.

classifiernn.Linear

Linear projection from the 1280-ch head (scaled by max(1.0, width_mult)) to config.num_classes.

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. Loss is the standard categorical cross-entropy, computed only when labels is not None.

Examples

Run inference on a batch of 224×224 RGB images:
>>> import lucid
>>> from lucid.models.vision.mobilenet_v2 import mobilenet_v2_cls
>>> model = mobilenet_v2_cls()
>>> x = lucid.randn(4, 3, 224, 224)
>>> out = model(x)
>>> out.logits.shape
(4, 1000)
Retarget to CIFAR-10:
>>> model = mobilenet_v2_cls(num_classes=10)
>>> model.config.num_classes
10

Used by 2

Constructors

dunder

init

→None

__init__(config: MobileNetV2Config)

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.mobilenet_v2 import mobilenet_v2_cls >>> model = mobilenet_v2_cls() >>> x = lucid.randn(4, 3, 224, 224) >>> out = model(x) >>> out.logits.shape (4, 1000) Retarget to CIFAR-10: >>> model = mobilenet_v2_cls(num_classes=10) >>> model.config.num_classes 10