data

ImageClassificationOutput

extendsModelOutput

ImageClassificationOutput(logits: Tensor, loss: Tensor | None = None, hidden_states: tuple[Tensor, ...] | None = None, attentions: tuple[Tensor, ...] | None = None)

source edit

Output of any {Family}ForImageClassification model.

Attributes

logitsTensor

Pre-softmax class scores, shape (B, num_classes).

loss(Tensor or None, optional)

Scalar cross-entropy loss when labels were supplied to forward; otherwise None.

hidden_states(tuple[Tensor, ...] or None, optional)

Per-layer feature maps when requested via output_hidden_states=True.

attentions(tuple[Tensor, ...] or None, optional)

Per-layer attention weights for transformer-family classifiers.

Notes

Returned by every classifier registered under AutoModelForImageClassification — CNN-family (ResNet, EfficientNet, …) and ViT-family backbones alike.

Examples

>>> import lucid
>>> from lucid.models import create_model
>>> model = create_model("resnet_18_cls", num_classes=10).eval()
>>> out = model(lucid.randn(1, 3, 224, 224))
>>> type(out).__name__
'ImageClassificationOutput'
>>> out.logits.shape
(1, 10)
>>> out.loss is None              # no labels were passed
True

Used by 26

… 14 more

Constructors

dunder

init

→None

__init__(logits: Tensor, loss: Tensor | None = None, hidden_states: tuple[Tensor, ...] | None = None, attentions: tuple[Tensor, ...] | None = None)

source edit

data

ImageClassificationOutput

extendsModelOutput

ImageClassificationOutput(logits: Tensor, loss: Tensor | None = None, hidden_states: tuple[Tensor, ...] | None = None, attentions: tuple[Tensor, ...] | None = None)

source edit

Output of any {Family}ForImageClassification model.

Attributes

logitsTensor

Pre-softmax class scores, shape (B, num_classes).

loss(Tensor or None, optional)

Scalar cross-entropy loss when labels were supplied to forward; otherwise None.

hidden_states(tuple[Tensor, ...] or None, optional)

Per-layer feature maps when requested via output_hidden_states=True.

attentions(tuple[Tensor, ...] or None, optional)

Per-layer attention weights for transformer-family classifiers.

Notes

Returned by every classifier registered under AutoModelForImageClassification — CNN-family (ResNet, EfficientNet, …) and ViT-family backbones alike.

Examples

>>> import lucid
>>> from lucid.models import create_model
>>> model = create_model("resnet_18_cls", num_classes=10).eval()
>>> out = model(lucid.randn(1, 3, 224, 224))
>>> type(out).__name__
'ImageClassificationOutput'
>>> out.logits.shape
(1, 10)
>>> out.loss is None              # no labels were passed
True

Used by 26

… 14 more

Constructors

dunder

init

→None

__init__(logits: Tensor, loss: Tensor | None = None, hidden_states: tuple[Tensor, ...] | None = None, attentions: tuple[Tensor, ...] | None = None)

source edit