ConvNeXt-Tiny backbone (Liu et al., 2022).
Builds the canonical ConvNeXt-T configuration: depths=(3, 3, 9, 3),
dims=(96, 192, 384, 768). The final stage produces a
feature map; forward_features
globally averages and LayerNorms to a vector.
Approximately 29M parameters — Swin-T equivalent.
Model Size
Parameters
pretrainedbool= FalseIf
True, loads ImageNet-1k pretrained weights when available
in the model zoo. Defaults to False.**overridesobject= {}Keyword overrides on top of the canonical ConvNeXt-T config.
Each override must match a field of
ConvNeXtConfig.Returns
ConvNeXtA ConvNeXt backbone returning a flat
feature.
Notes
ConvNeXt-T reaches 82.1% top-1 on ImageNet-1k (Liu et al., 2022, Table 1). See arXiv:2201.03545.
Examples
>>> import lucid
>>> from lucid.models.vision.convnext import convnext_tiny
>>> model = convnext_tiny()
>>> x = lucid.randn(1, 3, 224, 224)
>>> feat = model.forward_features(x)
>>> feat.shape
(1, 768)