inception_v4

→InceptionV441.1M

inception_v4(pretrained: bool = False, overrides: object = {})

source edit

Inception-v4 feature-extracting backbone.

Builds an InceptionV4 with the topology of Szegedy et al. 2017, Figure 9: multi-branch stem → 4× Inception-A → Reduction-A → 7× Inception-B → Reduction-B → 3× Inception-C, ending in an $8\times8\times1536$ feature map for a $299\times299$ input. 41.1 M parameters (the published classifier's 42.7 M less its 1536 → 1000 head).

Model Size

Parameters

pretrainedbool= False

No pretrained weights are published for the headless backbone; True raises NotImplementedError rather than returning a randomly initialised model. The ImageNet-1k checkpoint belongs to inception_v4_cls, which the error message names.

**overridesobject= {}

Keyword overrides forwarded into InceptionV4Config (e.g. in_channels).

Returns

InceptionV4

Backbone with the Inception-v4 configuration applied (or with overrides merged on top of it).

Notes

See Szegedy et al., "Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning", AAAI 2017. The paper defines a single network, so there are no size variants (H11).

Examples

>>> import lucid
>>> from lucid.models.vision.inception_v4 import inception_v4
>>> model = inception_v4().eval()
>>> out = model(lucid.randn(1, 3, 299, 299))
>>> out.last_hidden_state.shape
(1, 1536, 8, 8)

Used by 1

lucid.models.vision.inception_v4

>>> import lucid >>> from lucid.models.vision.inception_v4 import inception_v4 >>> model = inception_v4().eval() >>> out = model(lucid.randn(1, 3, 299, 299)) >>> out.last_hidden_state.shape (1, 1536, 8, 8)