pvt_v2_b5

lucid.models.pvt_v2_b5(img_size: int = 224, num_classes: int = 1000, in_channels: int = 3, **kwargs) → PVT_V2

The pvt_v2_b5 function instantiates the PVTv2-B5 model, the largest configuration in the Pyramid Vision Transformer V2 family. It significantly increases the depth of the third stage to 40 transformer blocks, delivering exceptional feature representation capacity suited for high-end vision tasks such as segmentation and detection on large-scale datasets.

Total Parameters: 82,882,984

Function Signature

@register_model
def pvt_v2_b5(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 for customization, including:

  • embed_dims (list[int]): Embedding dimensions across four stages. Default is [64, 128, 320, 512].

  • depths (list[int]): Number of transformer blocks per stage. Default is [3, 6, 40, 3].

  • num_heads (list[int]): Multi-head self-attention head counts per stage. Default is [1, 2, 5, 8].

  • mlp_ratios (list[int]): MLP expansion ratios per stage. Default is [8, 8, 4, 4].

  • sr_ratios (list[int]): Spatial reduction ratios for attention projection. Default is [8, 4, 2, 1].

  • qkv_bias (bool): Whether to include bias terms in query/key/value. Default is True.

  • norm_layer (Callable): Normalization layer used in the model. Default is nn.LayerNorm(eps=1e-6).

Returns

  • PVT_V2: An instance of the PVT_V2 class configured as a PVTv2-B5 vision transformer.

Examples

>>> import lucid.models as models
>>> model = models.pvt_v2_b5()
>>> 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, 6, 40, 3], sr_ratios=[8, 4, 2, 1], linear=False, ...)