VGG-13 feature-extracting backbone (config B, no BatchNorm).
Builds a VGG with the paper-cited VGG-13 topology:
per-block convolution counts (2, 2, 2, 2, 2). Approximately
133.0 M parameters in the full classifier variant. Reaches a top-5
ImageNet validation error of 9.9% (Simonyan & Zisserman, ICLR 2015,
Table 3, configuration B, single test scale S = Q = 256). 8.8% is
configuration D.
Model Size
Parameters
pretrainedbool= FalseNo checkpoint is published for
vgg_13, so True raises
NotImplementedError rather than return randomly initialised weights.
vgg_13_cls is the checkpointed variant.**overridesobject= {}Keyword overrides forwarded into
VGGConfig.Returns
VGGBackbone with the VGG-13 configuration applied (or with
overrides merged on top of it).
Notes
See Simonyan & Zisserman, ICLR 2015, Table 1 (configuration B).
Examples
>>> import lucid
>>> from lucid.models.vision.vgg import vgg_13
>>> model = vgg_13()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape # (B, 512, 7, 7)
(1, 512, 7, 7)