efficientnet_b7

→EfficientNet

efficientnet_b7(pretrained: bool = False, overrides: object = {})

source edit

EfficientNet-B7 feature-extracting backbone (compound coefficient $\phi = 6$ ).

Builds an EfficientNet with width multiplier 2.0 and depth multiplier 3.1 — the largest variant in the original paper. Approximately 66M parameters and 84.4% ImageNet-1k top-1 accuracy at the 600×600 native resolution (Tan & Le, 2019, Table 2), state-of-the-art at the time of publication.

Model Size

Parameters

pretrainedbool= False

Reserved for future pretrained-weight loading. Currently ignored.

**overridesobject= {}

Keyword overrides forwarded into EfficientNetConfig.

Returns

EfficientNet

Backbone with the B7 configuration applied (or with overrides merged on top of it).

Notes

See Tan & Le, "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks", ICML 2019 (arXiv:1905.11946), Table 2. Head expansion: round(1280 · 2.0) = 2560 channels. Memory footprint at 600×600 is substantial — plan accordingly for training.

Examples

>>> import lucid
>>> from lucid.models.vision.efficientnet import efficientnet_b7
>>> model = efficientnet_b7()
>>> x = lucid.randn(1, 3, 600, 600)
>>> out = model(x)
>>> out.last_hidden_state.shape
(1, 2560, 1, 1)

Used by 1

lucid.models.vision.efficientnet

>>> import lucid >>> from lucid.models.vision.efficientnet import efficientnet_b7 >>> model = efficientnet_b7() >>> x = lucid.randn(1, 3, 600, 600) >>> out = model(x) >>> out.last_hidden_state.shape (1, 2560, 1, 1)