pvt_v2_b2_li

lucid.models.pvt_v2_b2_li(img_size: int = 224, num_classes: int = 1000, in_channels: int = 3, **kwargs) → PVT_V2

The pvt_v2_b2_li function instantiates the linear-attention variant of the PVTv2-B2 model. This version leverages a lightweight attention mechanism, enabling reduced computational complexity while maintaining strong performance on vision tasks. The “li” suffix stands for linear, referring to the use of Linear Spatial Reduction Attention as proposed in the PVTv2 paper.

Linear spatial reduction attention improves efficiency by:

  • Computing attention in a reduced spatial domain.

  • Scaling better with input resolution by avoiding quadratic attention cost.

  • Maintaining global receptive fields with significantly fewer operations.

This variant is well-suited for deployment scenarios where compute or memory constraints are tight.

Total Parameters: 22,553,512

Function Signature

@register_model
def pvt_v2_b2_li(img_size: int = 224, num_classes: int = 1000, in_channels: int = 3, **kwargs) -> PVT_V2

Parameters

  • img_size (int, optional): The input image size. Default is 224.

  • num_classes (int, optional): The number of output classes for classification. Default is 1000.

  • in_channels (int, optional): Number of input channels. Default is 3.

  • kwargs (dict, optional): Additional parameters passed to pvt_v2_b2, with linear=True fixed.

Returns

  • PVT_V2: An instance of the PVT_V2 class configured as a linear-attention PVTv2-B2 model.

Examples

>>> import lucid.models as models
>>> model = models.pvt_v2_b2_li()
>>> print(model)
PVT_V2(img_size=224, num_classes=1000, patch_size=7, embed_dims=[64, 128, 320, 512],
       num_heads=[1, 2, 5, 8], depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], linear=True, ...)