VGG-19 with BatchNorm — feature-extracting backbone.
Builds a VGG with the VGG-19 topology
arch=(2, 2, 4, 4, 4) and lucid.nn.BatchNorm2d between
each Conv and its ReLU. 20.04 M parameters without the FC head (vgg_19_bn_cls adds it for 143.7 M). The
deepest VGG variant; BatchNorm is particularly helpful at this
depth.
Model Size
Parameters
pretrainedbool= FalseNo checkpoint is published for
vgg_19_bn, so True raises
NotImplementedError rather than return randomly initialised weights.
vgg_19_bn_cls is the checkpointed variant.**overridesobject= {}Keyword overrides forwarded into
VGGConfig.Returns
VGGBackbone with the VGG-19-BN configuration applied (or with
overrides merged on top of it).
Examples
>>> import lucid
>>> from lucid.models.vision.vgg import vgg_19_bn
>>> model = vgg_19_bn()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape
(1, 512, 7, 7)