pvt_v2_b0

→PVT

pvt_v2_b0(pretrained: bool = False, overrides: object = {})

source edit

PVT v2-B0 backbone (Wang et al., 2022).

Builds the smallest PVT v2 variant: embed_dims=(32, 64, 160, 256), depths=(2, 2, 2, 2), num_heads=(1, 2, 5, 8), sr_ratios=(8, 4, 2, 1). Approximately 3.7M parameters — targeted at mobile / edge deployments.

Model Size

Parameters

pretrainedbool= False

If True, loads ImageNet-1k pretrained weights when available in the model zoo. Defaults to False.

**overridesobject= {}

Keyword overrides on top of the canonical PVT v2-B0 config.

Returns

PVT

A PVT backbone returning a flat $(B, 256)$ feature.

Notes

PVT v2-B0 reaches 70.5% top-1 on ImageNet-1k at 224x224 (Wang et al., 2022, Table 1). See arXiv:2106.13797.

Examples

>>> import lucid
>>> from lucid.models.vision.pvt import pvt_v2_b0
>>> model = pvt_v2_b0()
>>> x = lucid.randn(1, 3, 224, 224)
>>> feat = model.forward_features(x)
>>> feat.shape
(1, 256)

Used by 1

lucid.models.vision.pvt