SE-ResNet-152 feature-extracting backbone (no classification head).
Builds an SENet with ResNet-152 topology
(_SEBottleneck blocks stacked [3, 8, 36, 3]) plus
squeeze-and-excitation modules. Approximately 66.8M
parameters and 78.43% ImageNet-1k top-1 accuracy in Hu et al.,
2018 (Table 2: 21.57% top-1 error) — the deepest SE-ResNet
variant in the original paper. (78.66% is the shipped
checkpoint's metric, recorded in _weights.py.)
Model Size
Parameters
pretrainedbool= FalseNo checkpoint is published for
se_resnet_152, so True raises
NotImplementedError rather than return randomly initialised weights.
se_resnet_152_cls is the checkpointed variant.**overridesobject= {}Keyword overrides forwarded into
SENetConfig.Returns
SENetBackbone with the SE-ResNet-152 configuration applied (or
with overrides merged on top of it).
Notes
See Hu et al., "Squeeze-and-Excitation Networks", CVPR 2018 (arXiv:1709.01507), Table 2.
Examples
>>> import lucid
>>> from lucid.models.vision.senet import se_resnet_152
>>> model = se_resnet_152()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape
(1, 2048, 7, 7)