PVT v2-B3 backbone (Wang et al., 2022).
Builds the canonical PVT v2-B3 configuration:
embed_dims=(64, 128, 320, 512), depths=(3, 4, 18, 3) (a
deeper stage 3 than B2). Approximately 45.2M parameters —
ResNet-101 class on accuracy/cost.
Model Size
Parameters
pretrainedbool= FalseIf
True, loads ImageNet-1k pretrained weights when
available. Defaults to False.**overridesobject= {}Keyword overrides on top of the canonical PVT v2-B3 config.
Returns
PVTA PVT backbone returning a flat
feature.
Notes
PVT v2-B3 reaches 83.1% top-1 on ImageNet-1k at 224x224 (Wang et al., 2022, Table 1).
Examples
>>> import lucid
>>> from lucid.models.vision.pvt import pvt_v2_b3
>>> model = pvt_v2_b3()
>>> x = lucid.randn(1, 3, 224, 224)
>>> model.forward_features(x).shape
(1, 512)