csp_resnext_50_32x4d¶

lucid.models.csp_resnext_50_32x4d(num_classes: int = 1000, split_ratio: float = 0.5, stem_channels: int = 64, **kwargs) → CSPNet¶

The csp_resnext_50_32x4d function constructs a CSPNet variant using ResNeXt blocks as its transformation units. This combines the split-transform-merge topology of CSPNet with the grouped convolution efficiency of ResNeXt, yielding strong accuracy with lower computational redundancy.

This configuration mimics the original ResNeXt-50 (32x4d) layout, partitioned with Cross Stage Partial (CSP) connections.

Total Parameters: 22,509,864

Function Signature¶

@register_model
def csp_resnext_50_32x4d(
    num_classes: int = 1000,
    split_ratio: float = 0.5,
    stem_channels: int = 64,
    **kwargs
) -> CSPNet

Parameters¶

num_classes (int, optional): Number of output classes for classification. Default is 1000.
split_ratio (float, optional): Ratio of input channels routed to the residual stack in each CSP block. Default is 0.5.
stem_channels (int, optional): Output channels for the stem convolution. Default is 64.
kwargs (dict, optional): Additional keyword arguments passed to CSPNet.

Returns¶

CSPNet: A CSPNet-based model using ResNeXt-50 32x4d block structure.

Examples¶

from lucid.models import csp_resnext_50_32x4d

model = csp_resnext_50_32x4d(num_classes=1000)
x = lucid.random.randn(1, 3, 224, 224)
y = model(x)
print(y.shape)  # torch.Size([1, 1000])