CSPResNet-50 backbone — ResNet-50 with CSP-wrapped stages.
The CSPNet recipe applied to the ResNet-50 trunk (Wang et al.,
CVPRW 2020). Returns the final feature map only; pair with a
classifier head via cspresnet_50_cls.
Reference: Wang et al., "CSPNet: A New Backbone that can Enhance Learning Capability of CNN", CVPRW 2020 (arXiv:1911.11929).
Model Size
Parameters
pretrainedbool= FalseNo checkpoint is published for
cspresnet_50, so True raises
NotImplementedError rather than return randomly initialised weights.
cspresnet_50_cls is the checkpointed variant.**overridesobject= {}Fields of
CSPNetConfig to change from the CSPResNet-50
defaults (e.g. in_channels=1).Examples
>>> import lucid
>>> from lucid.models.vision.cspnet import cspresnet_50
>>> model = cspresnet_50()
>>> out = model(lucid.randn(1, 3, 224, 224))
>>> out.last_hidden_state.shape
(1, 1024, 7, 7)