densenet_264

→DenseNet

densenet_264(pretrained: bool = False, overrides: object = {})

source edit

DenseNet-264 feature-extracting backbone (no classification head).

Builds a DenseNet with the paper-cited DenseNet-264 topology: per-block dense-layer counts (6, 12, 64, 48), growth rate $k = 32$ , initial conv channels 64. Approximately 33.3 M parameters total — the deepest DenseNet evaluated in the paper, reaching a top-1 ImageNet error of 22.2% (Huang et al., 2017, Table 2).

Model Size

Parameters

pretrainedbool= False

Reserved for future pretrained-weight loading. Currently ignored.

**overridesobject= {}

Keyword overrides forwarded into DenseNetConfig.

Returns

DenseNet

Backbone with the DenseNet-264 configuration applied (or with overrides merged on top of it).

Notes

See Huang et al., CVPR 2017, Table 1. Final pre-classifier feature width is 2688 channels.

Examples

>>> import lucid
>>> from lucid.models.vision.densenet import densenet_264
>>> model = densenet_264()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape   # (B, 2688, 1, 1)
(1, 2688, 1, 1)

Used by 1

lucid.models.vision.densenet

>>> import lucid >>> from lucid.models.vision.densenet import densenet_264 >>> model = densenet_264() >>> x = lucid.randn(1, 3, 224, 224) >>> out = model(x) >>> out.last_hidden_state.shape # (B, 2688, 1, 1) (1, 2688, 1, 1)